Julianne Weis worked at USAID for nearly eight years. When the agency effectively closed in 2025, she went to the Hill to start advocating for foreign aid. Julianne and Clara discuss:

• The current political situation of foreign aid

• Why messaging (still) matters

• The role of whistleblowers during the DOGE crisis

• The next USAID-like agency

This interview has been edited for length and clarity.

Clara Collier: You were at USAID for many years, and you also co-founded Aid on the Hill, which is an aid advocacy organization. Do you want to start by briefly talking about your USAID background?

Julianne Weis: Sure. So I am a global health researcher by training. I spent many years in various African countries doing global health research. I then worked at USAID in global health research, in one of the offices in the Global Health Bureau, focused on implementation science research — particularly behavioral science, actually social behavior change. I worked at USAID for about seven and a half years.

In January 2025, we were effectively all laid off. They did keep a really small group of essential staff that they used to essentially close down USAID through July 1, 2025. And then there was another group that went on to the end of 2025, but that was truly a few dozen people from 15,000 of us worldwide.

The more important part was that all of our programs were paused in January 2025. They sent stop-work orders to every single aid project around the world, and in that moment, I just had complete, true panic about what would happen if they succeeded. I started going to the Hill to advocate for restoration of the life-saving projects. And the momentum just grew and grew and grew. We really filled a gap and a need — an information gap. That was the main thing: Congress was just getting no information from the State Department about what was actually happening. We were this really key sort of watchdog player at that time. And so it then became Aid on the Hill, which is now a fully fledged advocacy organization — funded and everything, but I don’t work there anymore.

Clara: One reason I really wanted to talk to you is that I hear a lot of conversations about, how should we reconstruct foreign aid — what should it look like what should we do — and much less on the politics. But of course, as we all learned last year, foreign aid is a political question. So what do you see as the political situation with foreign aid right now? Who supports it, who’s against it, which constituencies actually matter for making sure that there is robust political support for aid in the U.S.?

Julianne: I think it’s a way more hopeful story than people think. I think the narrative that foreign aid is a completely politically toxic area just does not bear out in the data at all.

There’s been a lot in the last year — a lot of focus group and survey data digging into this issue, including with Republicans in red states. Surveys and focus groups, as any researcher knows, can be very biased, but when the story is presented as, hey, this is what America is capable of doing, this is what foreign aid can achieve — there was just data that came out about the enormous bipartisan support for PEPFAR, for example. Life-saving support, really effective programs, things that truly make a difference, that don’t cost a lot of money in comparison to the U.S. budget. When that story is told, there’s an enormous amount of support for it. I think the problem is that when they attacked USAID, they had to make the case for that. And so they absolutely flooded media with misinformation about aid and about waste, fraud, and abuse.

All these lies about what aid was doing were just completely flooded across the media. And so people assumed everybody was buying into this narrative. I do think there’s probably a 30% of the American public that will never be behind this. But it is not at all the majority. Even among Republicans, it’s not a clear partisan line. There are actually people on the far left who are also very skeptical and dismissive of USAID. In the DSA — the Democratic Socialists of America — platform, they called for the abolishing of USAID and talked about it as an imperial, CIA-front kind of thing. So it’s not a straightforward “right versus left” or “Democrat versus Republican.” It’s a bit more complicated than that.

But overall, what we’ve seen from a lot of the polling and survey data, and from doing advocacy and activism, is that the conversation of America being a good player on the world stage and doing really effective foreign assistance is actually a very compelling argument, a compelling policy point, that rallies a lot of different kinds of people.

The PR of foreign aid

Clara: Do you think the aid and development community was failing to communicate this previously?

Julianne: Yes, absolutely. I do agree with that postmortem, that aid was really bad at this. Yes, I do agree.

Clara: What were the biggest mistakes?

Julianne: Okay, so foreign aid is inherently very complex in a lot of ways. But I think we overcomplicate it from a communications point of view. There’s a lot of apology about aid, because there are inherent power dynamics, there’s neocolonialism, there are problematic elements of foreign assistance. We don’t want to go back to like 1985 Live Aid, pretending or assuming that all people in African countries are just desperate and starving. There are ways that we need to overcome those narratives.

But the biggest mistake was just not talking about the huge successes. Global health has been enormously successful. The reductions in child mortality in the last 20 years are astonishing. That is a huge success story. And when people understand that we were actually making an impact, they’re way more likely to support it. It’s kind of like climate change as well: If people only hear things that are bad or hopeless, then they just disengage entirely. But if they know that there are ways that you can actually have a huge impact — there’s a malaria vaccine, we’re coming out with a new TB vaccine, we’ve had huge strides in global health — leaning into those positive narratives and the impact would serve us a lot better than being too apologetic, too hidden, too reductionist, or being like, “It’s too complicated to tell this story.”

Clara: When we were doing a lot of PEPFAR coverage last year, people who were more PEPFAR-skeptical would often say, “Well, is it supposed to last forever? Why are we doing this if there’s no progress?” And you’d have to explain, “No, no, no, no. There’s been a lot of progress.”

Julianne: Yes. I was actually on a webinar this week, and Samantha Power mentioned that. She said a big mistake was that we didn’t communicate that aid was a graduation model.

Clara: Of course, it’s also the case that implementing that graduation model is hard.

Julianne: Yes. And there’s just so much that’s misunderstood. I have family that consume a lot of very right-wing media. And my mom sent me this meme. She’s like, “I don’t agree with this at all, Julianne. I just want you to see what’s out there and what people are seeing.” And it was a meme of a picture of an African wattle-and-stick home. And it said “Africa pre-aid.” And then it was another picture of the same exact wattle-and-stick home: “Africa, this many trillion dollars later.”

So reductionist. Just the idea that we’ve thrown all this money away, nothing has happened. And that’s really, really untrue. So it’s hard, because you want to make the case for aid and talk about the problems that are there — like, “this is a big problem.” There are still huge problems. Children are dying of diarrhea.

There are tremendous water and sanitation issues. But I think focusing so much on the problems has not served us well, in the sense that people have all these excuses — “oh, that’s not our problem to solve,” or “well, we’ve given all this money already,” all these different kinds of things. Whereas if we talk about, “hey, we did this and this is the outcome, and look how cool” — like, oh, now South Africa is doing this or whatever — then it helps people realize, we’re all in this together. There’s no perfect communication strategy.

I’m not going to say it was all aid’s fault that we had to communicate it — “if we had communicated better, Elon Musk wouldn’t have done this” — because I think he would have done it anyway. He had a clear vendetta and clear intention to tear down the U.S. government. So I’m not going to do victim-blaming. But yes, we definitely did that.

Does Congress matter?

Clara: This actually relates to my next question, which is: There’s the issue of how to build public support for aid. But then you have to ask, does that actually influence Congress? And now on top of that we also have to wonder, does Congress matter?

I think you’re right in saying that Musk was probably going to try to do what he did anyway. He was already convinced by all these conspiracy theories. But if there had been much more public support for aid, what difference do you think that would have made in the moment? Would other actors have responded differently? And should that influence the strategy going forward? Maybe we shouldn’t even think of that as a repeatable thing you can plan for.

Julianne: That’s really hard. I don’t think there was anything we could have done differently, because I heard from a Republican senator’s chief of staff that in those early months, USAID was the number one constituent-call issue from their red state.

Clara: In favor or against?

Julianne: In favor. Saying, “Why are you cutting foreign aid? You’re killing people.” That was in large part — this I can say — it was Utah. That was the state. It was because of the tremendous advocacy work of groups like Mormon Women for Ethical Government, who were sounding the alarm.

So I know that there were lots of constituent calls. I know there was a lot of media attention. Republicans at that stage — Trump had just come to office and they just were not going to publicly push back against him. They just were not going to do that. What they did instead — and this is what they all told us and assured us, and it did end up happening — is they restored the budget for the following year. They restored the funding, and they put all these checks on the State Department. But they had decided they were going to sacrifice USAID — let them have USAID, we’ll continue to do PEPFAR under the State Department, we’ll still do all the good life-saving work, and we’ll restore that, don’t worry, but we can’t step in and stop this.

And I’m like, it’s really frustrating, because the State Department is just not staffed up. They’re not equipped to do aid at the level that is necessary. There have been tremendous implementation issues, and there will continue to be. So I think there was a naïveté, but I do think Congress did as much as they were willing to do as a Republican Congress, in the first year of a new administration.

If Elon Musk had gone after USAID later, it might have been different. Look at the Department of Education, which actually still exists. That was one they said they were going to demolish. It’s actually still there. They’ve kind of been like, “Oh, never mind.” I mean, it’s been greatly reduced. There have been humongous RIFs across all of the federal government. But I think the timing of it meant there was really nothing they were going to do, because it was so early in the game. If Musk and DOGE had gone after USAID later, they would have pushed back more. But there just wasn’t much time.

Does constituent voice matter to Congress — yes, 100%. It very much does. Does anything Congress does matter? In normal times, it matters a lot. In this administration, it doesn’t matter very much, because the executive branch is just kind of doing whatever they want regardless of congressional intent. And that’s not just with foreign aid. Look at NIH, look at NSF — Congress has restored all these budget lines. They’ve even demanded that, you know, you have to fund this program, you need to hire staff, you need to do these things. Congress is actually pushing back, and that is because of constituent voices, that is because of advocacy. But the executive branch is completely rogue, does not care about the law. That’s really difficult.

Clara: So where does this leave aid advocates? Is the goal to build enduring support for the future when the administration is less insane? Should the focus be more on figuring out checks on executive power? What does it make sense to focus on?

Julianne: I still think it makes a lot of sense to focus on the immediate — on staving off the worst harms. There’s too much talk about the future of foreign aid that is ignoring what’s happening right now. And you still need to be engaged with what’s happening right now, because Congress, again, they did appropriate all this money. The State Department has all this money that they should be spending, that they could be spending. They are spending some of it. They are doing global health work. So it’s really important to still be engaged in the immediate, to ensure that at least something passable is happening, that HIV treatment continues, that humanitarian aid continues.

In the longer term, it’s hard for me to think about, how do you create executive-branch checks that will actually hold? Technically we have them in our Constitution and in our law. They just don’t care. What do you do when you have absolutely shameless judges? It’s an authoritarian takeover. I don’t know if you can legislation-proof this kind of thing.

USAID does still legally exist. It’s still there, because Congress would have had to pass a law to abolish it, which they haven’t. But it’s been effectively shut down just by all the money being taken away, the staff being taken away, the power of it being taken away.

What makes the administration act?

Clara: I do want to push, though. What is the role of advocacy in this situation? One thing you said earlier was that when USAID was actually being demolished, there was a very active attempt to control the media narrative around it, which suggests that on some level they do feel constrained by public opinion. They do care.

Julianne: Yes. That’s absolutely true. And I do think that’s why DOGE stopped. That’s because it was so deeply unpopular. That’s why DOGE went away. That’s why they didn’t demolish the Department of Education. So they are absolutely constrained by public attention.

They were trying to get the narrative out in front, saying that USAID was only wasteful and that all the reports of deaths from aid cuts are a lie and all that. So that’s why I said, focus on things that are happening now, to ensure that they’re still programming the money, they’re still getting money out the door, they’re still supporting essential services that the U.S. has committed to. And the way to do that is by continuing that media spotlight — because you’re right, they can be shamed. When Nicholas Enrich at USAID put those memos out in March, April, whatever, in 2025, that completely changed the game, because he proved —

Clara: Explain this incident, for people who may be less familiar with it.

Julianne: So when they had done the stop-work order, they said that they would have a waiver for life-saving assistance. They had put the majority of USAID staff on administrative leave, or fired all the contractors. There was a core small group of essential employees that were supposed to be continuing to implement the waivers and shut down USAID.

Nicholas Enrich was one of them. And he put out a memo to the public. He sent it to the New York Times and he sent it to different media, documenting very, very clearly that the waiver system was not working, that no aid was happening, that everything was still stopped, that the response to Ebola in Uganda was completely stopped, that all humanitarian aid was stopped, that nothing was moving forward.

Up until that point, Rubio, the State Department, Pete Marocco, all these people had been saying, “Don’t listen to NGOs. We’re not stopping life-saving aid. We’re still doing life-saving aid. Everything is fine.” And Nick Enrich’s memos demonstrated very publicly that that was a lie. He was a whistleblower from inside USAID.

We had been on the Hill at that point doing advocacy for a while, and Congress was pushing State Department. They were like, “Okay, now show us, what are the projects that are active?” They were trying, but the State Department was just ignoring Congress entirely. But after the media reports from the memos, all of a sudden the State Department turned on a bunch of life-saving programs again.

That was the ticket that actually restored the most essential programs. It was still very rocky in the months after that. It was not immediate. It was still really difficult. But it was because of the media attention and shaming. So in this right moment, that still matters, because it’s still very messy. You need continuous advocacy — that public shaming and advocacy, dual-hat, to make sure that — you’re seeing this around the global health MOUs, the memoranda of understanding.

There is a lot of press about those. There was that Stephanie Nolen article in the New York Times about HIV coming back in parts of Zambia. That kind of reporting really, really does anger the administration, and it will force them to act. So yeah, maybe it is more positive — the idea that the executive branch can be held to account.

Will this ever, could this ever happen again? It seems, in my view, unlikely. But there’s no law that you could write that would prevent this from happening, because there are already laws that should be preventing this from happening, and they did. I think it will always be the role of active civic engagement.

Clara: I should say, I’m not sure how optimistic I am about public shaming, but I think it’s important to be realistic about what levers actually exist.

Julianne: I completely agree with you. It’s a big reason why I have spoken to media for this whole year. From the beginning, I was sending things to media, talking to media, trying to get other colleagues to talk to media. I think that’s extremely important when you see illegal things happening, when you see things that are extremely dangerous and going to lead to the loss of lives. That’s when you have to sound the alarm publicly, and not just try to work from the inside. Working from the inside is not enough.

Clara: Do you think that was an issue early on — people were too reluctant to go to media?

Julianne: Yes, 100%, and it still is. There are a lot of shenanigans still going down. It’s hard. I understand people have jobs and livelihoods and they don’t want to lose them.

And there really is this tug-and-pull of, “Can I do more good from the inside?” I think it’s a mix. It’s not cut and dry. I know there are a lot of people doing really good things on the inside. But when Nick published his memos, there were a lot of USAID people who were angry at him that he did that. And I think that’s bananas, because he was clearly really effective. But a lot of people were like, “No, we just have to keep working from the inside.” That’s a very civil servant attitude. But it’s just not normal times. These things are not normal.

An independent aid agency

Clara: One thing I know that Aid on the Hill, at least, has advocated for is the importance of having an independent aid agency. Right now, aid runs through the State Department. And this is not unusual internationally — many countries do operate this way. Why do you think it’s important to restore USAID or some other independent aid agency outside of State? And what kind of appetite there is for that right now?

Julianne: Yes, I do think it’s better served independently. Look at happened with, say, DFID turning into FCDO in the UK — and Canada, Australia — those are the three examples always cited of when an independent agency was folded into the foreign office. We can see when this happens that diplomacy suffers in some ways, and also the quality of foreign assistance suffers. It becomes much too entangled. They are separate skills, they are separate expertise, they are separate intentions. A person that is a diplomat is serving a country’s interests and trying to work with that country to further America’s interests.

That is actually different than if your goal is, say, reducing malaria deaths in this province. And while foreign aid is of course related to soft power, it’s related to foreign policy, there are ways that an independent agency can operate more neutrally. Even if it’s tied with America, it does have an air of neutrality. Even the idea that the State Department now wants to do all aid branded just with the American flag — that can be really risky. There are a lot of places where that can really backfire. It’s not actually a good idea, from a development perspective or a humanitarian perspective.

When you fold development inside a diplomacy apparatus, it’s just never going to have the expertise or the impetus. The reason for it is not going to be, we are reducing poverty, we are improving global health. It’s always going to be much more tied to America’s goals. And that’s what the Trump administration wants, because it’s all America First. It’s all about the country and about America’s benefit.

Clara: I want to play devil’s advocate for a moment here, because I think a lot of people would say, what’s wrong with that? Why shouldn’t aid serve America’s goals?

Julianne: Maybe this is my politics coming out.

Clara: I know, but I think it’s important to articulate a response.

Julianne: You’re right. Well, in my opinion, it’s in America’s interest to have other countries be developed and thriving and providing services for their populations. It’s in America’s interest to live in a safe, prosperous world. That is good. But the way to ensure that countries achieve those goals of development are not always explicitly tied to a U.S. company, for example.

If our intention is first benefiting America, that can actually backfire and not be good for that country, and then we’re not actually achieving that development goal. I just think it’s in America’s interest that the DRC and Uganda have their own national epidemic-control apparatus. That is good for America, because that means we are not dealing with outbreaks in America — we are controlling outbreaks at the source. And not “we,” but the countries themselves are doing that. If that always has to be tied back to an American company, an American CDC, an American whatever, that’s really limiting the tools in our toolbox. And it’s really limiting how we’re partnering with that country. The country could lose trust with us. They could say, “You don’t actually care about us. We’re going to go work with somebody else. We don’t want to work with you.” Whereas America used to be seen as a true partner.

America had a very different position than China on the world stage. China was giving loans. China is not giving aid — if they’re giving loans, they’re bringing Chinese workers in to build a road. America would work with the country, do government to government, just full-on give the funding directly, work with partners like they build a hospital. It would be no-strings-attached. It is really difficult to quantify how important that is, and how we have now ruined that. But I don’t think it’s ruined forever. I talked to a lot of people from different countries — just this week, I talked to a colleague who’s from the Philippines, and he’s like, “Do you think USAID is going to come back? I really hope USAID comes back,” and he has some story about USAID and the Philippines doing this or that. And again, what USAID was doing was seen by the population as something benefiting them and building them up, and not just, “Oh, they’re doing this to get something for America” — which is how aid is being framed now.

That’s a very long answer, but I really, really do believe it’s better served as an independent aid agency. In terms of, is there appetite for that — it’s to be determined. There are a lot of people who are like, “Oh no, it’s never going to come back.” Other people are too optimistic — like, “Yeah, they’re going to restore USAID.” It’s not USAID. I think that’s important. If they want to just build out the Millennium Challenge Corporation or something, I don’t know, that’s fine. I just think it’s really important to have expertise in international development, and economics, and global health, and epidemiology. These are clear skill sets. USAID had all of these. It was the most highly educated federal agency — more than NASA, more than NIH — because these are actual skills of how you do development. It’s complex. It’s difficult. It’s something you can’t just wave away. It’s a clear skill set that should be fostered in an agency.

The next USAID

Clara: These people obviously are still there. What would it take to make that happen?

Julianne: I think a lot of people would come back. Absolutely. I personally probably wouldn’t. There’s a lot of trauma from what happened last year. It’s difficult. I think it would almost feel like a sacrifice. It would feel very scary, because you’d just always be waiting for the shoe to drop again. But there are a lot of young professionals going into the fields of international development, and I think they could do it. It doesn’t have to be bigger. It could be smaller. It could be more targeted. I think that’s fine. But there absolutely has to be something restored.

Clara: USAID was huge. It did a million different things. If there’s a new USAID-like organization, how should it be targeted? Should it be focusing on health and life-saving aid?

Julianne: I don’t think it should just be health and life-saving aid, because so much of global health problems are actually governance problems. So I do think it’s important to have really strong support for governance-type programming, like taxation — those kinds of really boring things.

Clara: How was that handled before?

Julianne: It was a huge part of USAID’s work. I think that’s really critical. I do think it could be more targeted. There were always new things being added, and never anything really taken away. It could be better designed for what countries actually need and want, and not always what DC wanted. There are ways it could be more flexible in how the aid projects were designed, and who was setting the agendas. I mean, everybody who worked at USAID would have said it was absolutely due for reform. It was not in any way perfect. It should never be restored as it was, exactly. If this is the chance to build something better, yes, let’s do that.

This is related to one of the problems with PEPFAR — PEPFAR has been enormously successful, but PEPFAR was such a vertical program at the outset, and it created essentially parallel health systems, which is just not ideal. And I understand why it did that. Again, it was very effective. But in the last years, it was really shifting to more of a health-systems approach, and having HIV be much more integrated into primary healthcare. So there are definitely ways it could be better.

Clara: You said something about how programs could be more focused on what partner-country governments need than what DC needs. How do you do that structurally? Because it seems like everybody agrees, and everybody has agreed for a long time, that this would be good. But it keeps not happening. It seems that it’s just very difficult to arrange those incentives for an organization that ultimately is based in DC and hires people in DC and is accountable to American political pressure. What do you think can be done on a structural level to combat that tendency?

Julianne: I always said the DC offices for USAID were way too bloated. There was so much — I worked in Washington DC, but a huge amount of my time, most of my time, was actually supporting the country offices, supporting the missions. Because they were super, super understaffed. So I think a great way to make things more country-led is to just completely decentralize it and have the missions — like, have the Kenya office — have much more control over the budget, and the planning, and everything.

So they’re not always responding to Washington requests, but really having things be country-led, and then regionally. And I would say, restore the foreign service national staff. Foreign service nationals were the institutional knowledge. They were there for many, many years. They would stay in an embassy working on development, having the relationships with governments, but they were not empowered as leaders. Foreign service officers would be cycling through every two to three years, and then they would be the ones —

Clara: Just to make sure I understand this, can you explain this role in a USAID mission?

Julianne: For example, in Malawi, the USAID mission in Malawi was 90% foreign service nationals. Those are people from Malawi that worked for USAID. They worked there for years — some for 20 years.

The majority of the workforce of USAID were foreign service nationals — people from the countries. I guess people don’t realize that. These are extremely educated, incredible people, that know the country, know the government. But USAID had this legacy colonial structure, where the mission directors and the health office directors and all top leadership were still American. They were foreign service officers who would cycle through on different tours. They’d go to Kenya for three years, then to Malawi for three years, they do these different tours, and they were the ones that would make the decisions.

There were a lot of amazing FSOs that were good at listening to local FSNs and local partners. But there were a lot who were not good, being very honest here. If I were to resurrect USAID and do something different, I would say hire FSNs back and give them much more power to decide what works with their partner government, with their civil society partners, with their communities. Give much more of that power locally. That’s what I would do.

Clara: We’re coming up on time. Is there anything else that you really wish people understood about aid and aid advocacy, that you feel they’re getting wrong?

Julianne: That it’s not hopeless, and that the story is not over. They restored the funding. There is humongous will in Congress to do something better. There really is. I know people don’t believe that, but it’s true — they do actually want to get this right. There is an intention to fix what happened. There really, really is within Congress. The executive branch is a different story. But Congress is listening, and they are worth engaging.

And whenever we talk about aid and aid effectiveness — you just cannot beat the scale of working at a government-to-government level, at a U.S. government level. They just have a different positionality in these different countries as a partner than any NGO or philanthropy. They will never be able to operate at the scale of government. So it’s really critical to get that government piece correct — better, more effective. It should not just be ignored. We can’t just work around it.

It’s difficult, it’s challenging, I know it’s frustrating — but we really have to contend with it, or else we’re just going to be doing side projects and then ignoring these enormous governance challenges that continue to play out. A lot of development and aid issues come back to issues of governance, and issues of government capacity, and state capacity, and those kinds of institutional, structural challenges that we keep facing. So we have to deal with them.

Spend enough time in online spaces, and you may come across a compelling — and counterintuitive — narrative: Sexism and repressive gender norms aren’t to blame for women being underrepresented in STEM fields relative to their population share. The real reason, or so this narrative goes, is that women are finally free not to enter STEM fields at all.

This account stems from a 2018 study in which the authors measured indicators of women’s science participation against the World Economic Forum’s Global Gender Gap Index. The headline finding is the so-called gender equality paradox, or the claim that women in more gender-equal countries are less likely to graduate with STEM degrees or pursue more traditionally masculine careers — and more likely to have typically feminine interests, personalities, and hobbies.

The authors propose an explanation: In more gender-equal countries, higher security and life satisfaction mean women are free to choose the careers they really want over high-paying technical ones. In this reading, efforts to increase women’s participation in STEM run counter to their revealed preferences for feminine careers, such as in education, carework, or administration. This view rapidly garnered traction in the popular discourse, becoming a talking point of both conservatives like Jordan Peterson and liberal writers like Noah Smith and receiving generally favourable coverage in mainstream outlets such as The Atlantic.

For individuals weary of feminist admonishments, the gender equality paradox was an effective counterpoint. But this popular success has outrun serious scrutiny of the methodology of the original paper.

Subscribe now

The problem of measurement

The gender-equality paradox appears to be highly sensitive to measurement choices. One group of researchers re-analyzed the paradoxical result using a different measure of gender equality. Instead of the GGGI, which captures relative gender gaps and female political representation, they used a simplified measure: the Basic Index of Gender Inequality, itself first proposed by the authors of the original 2018 gender equality paradox study.

This is a common way to test the robustness of a scientific finding, since credible results should not be dependent on slight differences in measurement. When researchers analyzed the 2018 dataset using BIGI, they found no relationship between gender equality and the percentage of post-secondary STEM graduates who were women. A modest relationship did reappear when BIGI was measured against the ‘propensity’ (as defined by the 2018 paper) of women to graduate with STEM degrees, but it was substantially weaker than the original estimate and existed only in a newer sample.

Even in the original study, using the more favourable ‘propensity’ metric, gender equality as measured by GGGI accounts for just over one-fifth of the variation in the likelihood of graduating with a STEM degree across countries — an insufficient correlation to suggest a causal relationship without further evidence.

The wider evidence shrinks too

There have been attempts to replicate the gender equality paradox in other domains, such as personality. In these cases, the scientists are testing whether more gender-equal countries have larger male-female gaps in common psychological indicators, like the Big Five personality traits. These studies typically find larger psychological sex differences in more equal countries.

But the methods used to arrive at this result expose another problem with this line of research. These studies rely on difference scores, created by subtracting the male mean from the female mean within each country and comparing that with measures of gender equality. This makes simple causal explanations suspect because the same difference score can arise from radically different conditions. Male math scores could be improving at twice the rate of female ones, or female math scores could be declining while male scores hold steady. The difference score would be the same in both scenarios. If the scores are the result of altogether different trends, they’re unlikely to reflect the same phenomenon.

A 2024 study from the University of Helsinki applied this critique systemically to the gender equality paradox. The authors re-analyzed multiple major studies exploring the paradox and looked at the underlying relationship between each variable – such as science attitudes and personality traits – and gender equality. Across all variables reanalyzed, there was no consistent pattern between them. Critically for the paradox, none showed men and women moving in opposite directions as equality improves.

Another problem: Men and women tend to move together statistically. If education, height, scientific literacy, or personality traits are higher for the men in country A than the men in country B, the women of country A will usually be higher in those traits compared to the women of country B as well — even if the gap between men and women in country A is also larger.

This can create major distortions if not handled carefully. When the averages for men and women are so strongly correlated, calculating a simple difference will remove all of their shared variance. This leaves only the small slice where the sexes diverge. That difference is then standardized by dividing it by what variation remains — and because so much of it was stripped out, the denominator used to calculate the final result collapses. This artificially inflates the apparent relationship and makes it look much bigger than it actually is.

The Helsinki researchers also noted that the studies they re-analyzed had strongly (in some cases, nearly perfectly) correlated means that were not sufficiently accounted for. They reevaluated the original results using more advanced statistical methods that standardized the scale of the variables and broke the difference scores into their core components. They found that the paradox’s magnitude had been overestimated by about 60% on average across studies. For the original 2018 study, the corrected effect for science-related attitudes was nearly half of what was originally reported.

The paradox that wasn’t

The results of the Helsinki study strongly imply, as the authors note, that the gender-equality paradox depends heavily on countries’ baseline differences distorting how the numbers are put together. Intuitively, this is because differences between nations are much greater than the variation between the men and women within a country. Thus, as the standard deviation is compressed and magnifies small sex differences, the resulting pattern will tend to mirror these country-based trends as it is inflated by their larger variation.

Evidence for this implication is provided by more recent research. This 2025 paper makes the key observation that gender roles and behaviours are not a single phenomenon across the globe. Rather, they are highly dependent on cultural context. Instead of treating all countries as one uniform group, the authors divided the world into seven cultural clusters — the Germanic / Protestant west, the Catholic west, Latin America, Orthodox Europe, the Muslim Middle East, Sub-Saharan Africa, and Asia — to control for differences in cultural context

The researchers analyzed multiple major studies published on the gender equality paradox using the same data and variables, but this time they controlled for culture. In practical terms, this means that differences in baseline characteristics — such as laws, cultural norms, or history — were held constant.

This matters mathematically. Clusters like this exist when groups of data points share unmeasured factors that affect the outcome. Ignoring them can be misleading, as the apparent differences may reflect the different cluster baselines, instead of the variable of interest.

One consequence of ignoring these patterns is a well-known statistical phenomenon known as Simpson’s paradox: a statistical quirk where a trend can appear to hold for an entire group, only to vanish or reverse once you break that group into the relevant sub-groups.

Imagine a study alleging class-based discrimination at a prestigious university. In this study, the poorer a student is, the less likely they are to be accepted into a program at the school. In effect, there is a correlation between family income and acceptance rates. But careful readers will notice a problem. Poorer students often have lower baseline test scores and GPAs. If you break the data down by GPA and SAT percentile, the correlation between income and acceptance could disappear or even reverse. In other words, the overall trend is likely a statistical artifact, created by combining students with very different academic profiles.

Researchers can address this problem by restricting analysis to countries that instead share a common baseline, or “intercept,” so that we don’t erroneously compare countries with very different cultural profiles without controlling for confounders. When the 2025 study accounted for this — so that Sweden was compared to Norway rather than to Iran — the gender equality paradox became nonsignificant in six of the datasets and reversed direction in one.

There was still finer analysis: The Germanic/Protestant West was subdivided into five language regions and the analysis was re-performed within that cluster. For the three studies on personality, all had large negative correlations between gender differences and equality, meaning that gender equality led to fewer gender-based personality differences. For the remaining four, the results were again nonsignificant.

The correlation between gender equality and science jobs on a global level seems to have been driven by Western countries, especially Germanic-speaking Protestant ones. Western countries, it seems, have unusually large gender gaps in domains such as STEM representation compared to the rest of the world. Across the board, whether a country was Western had a better correlation with gender gaps than gender equality did. Tellingly, invented predictors that correlated with Western-ness, such as colder average temperatures, also predicted gender gaps about as well as gender equality did. In a sense, the Western countries represented a group of outliers that the “paradox” could not live without.

This makes sense given we know that WEIRD —– that is, Western, Educated, Industrialized, Rich, and Democratic —– cultures are frequently outliers when it comes to factors like personality, cognition, decision-making, and, critically, gender dynamics. For precisely this reason, in psychology, there has been a growing push against extrapolating findings from such WEIRD countries to the rest of the world.

Various theories could explain why the Western nations have larger gender gaps to begin with. One is simply a matter of data quality. Psychological data obtained from Western populations is typically more internally consistent than data obtained from non-Western populations, meaning that the items on a measurement scale correlate more reliably with one another — possibly a result of these metrics generally being designed and fine-tuned in the West. As a result of greater reliability in one cultural context, this data will generally be better at detecting differences between people in that context, both between men and women and within each sex. In the 2025 study, the authors also found that data quality indicators predicted national gender differences as well as or better than measures of gender equality. When data quality alone was included as a control, the relationship between gender gaps and equality was reduced by 80% on average.

Ultimately, if the paradox reflected a real, causal relationship between gender equality and gender differences in STEM participation, vocation choice, and personality, controlling for country baselines should not cause the relationship to consistently collapse. What remains is a much narrower observation — that a specific set of Western, Protestant, industrialized countries have unusually large gender gaps in STEM, personality, interests, and other metrics — which is not really a gender equality paradox. It is a fact about a dozen or so countries in need of its own explanation.

What actually happens as gender equality rises

Careful readers may still need more convincing. After all, absence of evidence is not evidence of absence. But there is a powerful reason to doubt that gender equality causes gender gaps in STEM professions and elsewhere. Countries around the world have grown substantially more gender-equal over time, and single-country trends over time are not as exposed to the same statistical artifacts discussed above. If one really caused the other, we would expect to see at least some evidence that gender gaps grow in a given country as equality improves, and we wouldn’t need to torture cross-sectional OECD data to see a relationship.

But we see the opposite of what the paradox predicts: Where gender equality has any effect at all, it tends to make men and women more alike, not less.

A 2020 study attempted to replicate the gender equality paradox for psychological values over 14 years of data, rather than simply using cross-country correlations for a single year. This is a stronger design: Cross-sectional correlations across countries conflate gender equality with every other way countries differ, while longitudinal analysis like this tracks changes within each country over time, letting each nation serve as its own control.

The researchers studied 32 European nations from 2002 to 2016, and found that rising gender equality did not produce the widening differences the paradox predicts — the relationship was either null or ran the opposite direction. Another study published in 2023 did the same, this time focusing on the career aspirations of boys and girls. The researchers looked at a sample of European countries from 2006 to 2018, and found that as national equality and women’s empowerment improved, the career aspirations of male and female students converged. This effect can also be seen around the world in a number of domains. In the United States, the percentage of women in the STEM workforce has grown dramatically over the last half-century of feminist advancements.

The available evidence suggests that gender equality does not cause larger gender gaps. When gender equality improves in a given country, these gaps typically shrink. Rather, a handful of Western countries are both very egalitarian and report larger baseline sex differences across STEM aspirations, occupational choice, and a number of psychological traits.

It is only when researchers rely on highly questionable cross-country correlations, which are often biased by underlying cultural and socio-economic differences, that we can convince ourselves of any other story. But this is not good science. Correlation is not causation, and post-hoc, just-so stories are not good theories.

Gabrielle Sorensen is an economic researcher and freelance writer, with previous bylines in publications such as The Hub and Policy Options. She holds her M.A. in Economics from the University of British Columbia.

It’s been a year since the Trump administration dismantled USAID and severely disrupted the operations of PEPFAR, the global health program that prevented millions from dying of HIV and AIDS.

Dr. mike Reid served as PEPFAR’s Chief Science Officer from 2023 until this April. He’s seen all these changes and more play out from the inside. Clara and mike discuss:

• The current state of PEPFAR

• Why the biggest impacts of last year’s cuts are probably yet to come

• How (and how not) to transfer responsibility to partner governments

• What to expect from new AIDS drugs (and AI)

• The problems with the USAID, and what we should build instead

This interview has been edited for length and clarity.

Subscribe now

Clara Collier: I’m very excited to be talking to mike Reid. mike, you were the chief science officer of PEPFAR — the President’s Emergency Plan for AIDS Relief — until you left in April. So to start out, I want to establish: What does the chief science officer do? What was your job?

mike: My charge was to be the champion advancing scientific innovation within PEPFAR’s programs. So as new tools — whether those are therapeutic or diagnostic tools — came down the pipe, I would be the one responsible for consideration as to where they might fit in our portfolio.

In addition, historically, PEPFAR has funded a lot of research and science. As we’ve scaled programs, it’s been clear that there has been a lot of evidence gaps. So along the way, part of that role has involved commissioning and then translating science into programs. And then the third piece has always historically sought to be accountable to the broader scientific community. So I facilitated our engagement with our advisory board and the wider research community that is interested in or invests in HIV research in global settings.

Clara: This must have been such a strange year for you. On the one hand — and we’re going to talk about all of this in more detail — everything is getting ripped apart. And on the other hand, there’s are so many exciting medical advances, like lenacapavir — a new, highly effective antiretroviral therapy that lasts for six months. What was that like?

mike: I don’t want to overstate my own contribution here. Like many big organizations, there’s a whole team. We have experts in the particular technical domain, we have experts in engaging the pharmaceutical companies, and we have folks who are thinking about this from a country point of view.

That said, it’s hugely exciting. We’ve been tracking lenacapavir over the last three years through Phase III trials, and when those first results came out in November of 2024, particularly the PURPOSE 1 trial that showed 100% efficacy in adolescent girls and young women in sub-Saharan Africa, people were just so excited for the potential of this drug.

A lot of the work also involved engaging pharma and other stakeholders to think about how could we make it more affordable. What kind of volumes should the U.S. government be procuring that would shape the market so that generics would be able to enter at a competitive price? Those are all the things that people were thinking about, regardless of the change of administration. Obviously, that had a massive impact on how we were thinking about lenacapavir. And at the time of transition, it really wasn’t clear that the incoming administration had any appetite for lenacapavir. In fact, as you may know, they actually said “we want to put a pause on all prevention programming.” It took a fair amount of politicking and advocacy to persuade them that actually we needed to invest in prevention.

Clara: I know historically, PEPFAR has done a lot of market shaping for antiretroviral therapy and for other drugs. How has that been affected by what’s happened over the past year? What is PEPFAR doing for lenacapavir and some of these other new therapies now?

mike: There’s two questions there. In the very near term, I don’t think PEPFAR investment in market-shaping activities has changed from one administration to the next. With lenacapavir, there was a strategic pause, but then I think the incoming administration recognized that it also wanted to invest in len and saw the value of that. I think they were looking for a better deal for len than the Biden administration got from Gilead.

But from a 30,000-foot perspective, I don’t know that there has been a substantial shift in terms of the overall vision for using U.S. investments to drive down prices so that they’re affordable. One thing that the U.S. is moving towards is procuring all of its commodities through the Global Fund platform, through Wambo, rather than separately through Chemonics. That will allow procurement at even greater scale, and arguably help drive down prices even more. It may also increase inefficiencies, though, because Wambo is not nearly as efficient as Chemonics was. So that’s just one element of how these things are going to play out.

Clara: If the system is less efficient than Chemonics, why is there the expectation that switching would be an improvement? I’m trying to understand this.

mike: Chemonics is a private company. They take a big indirect. From a financial point of view, they offer a more expensive model of commodity procurement and distribution. But they also address many of the elements of the supply chain infrastructure that Wambo doesn’t. They facilitate addressing regulatory hurdles in country, they help support countries to actually forecast and then distribute drugs. Wambo is just a mechanism that procures drugs at scale. And so if the U.S. government decides that Wambo is its destination for commodity procurement, then they will also have to figure out: How do those drugs get distributed? How do you address the regulatory hurdles of introducing new tools?

That will probably require a thousand different solutions in a thousand different countries. And unless you’re going to use a single partner like Chemonics — which is not the desired trajectory — that’s how I think there may be inefficiencies, even though we’re moving away from a more expensive model.

PEPFAR today

Clara: This relates to another big thing I wanted to talk about. When I was preparing for this interview, I asked people who work in global health nonprofit what they were most curious about. The number one thing everybody wants to understand is: What is the current state of PEPFAR? What systems still exist? What stopped? What data sets are we still collecting?

We had this whole system for providing treatment and providing care and distributing these medications. What does this look like on the ground now? And I know this is a huge question, but is there a 1,000-foot overview of what is currently going on?

mike: Yes. There are still functions. It is beholden to the same legislative mandate that it was two years ago. Nonetheless, this administration has sought to do a couple of things that have had profound programmatic implications.

The first is that it has made a really rapid transition towards country ownership. That’s a good thing. Everybody in the PEPFAR space — almost everybody — is excited about the possibility that in the long term, HIV programs are funded, run, overseen by partner governments rather than the U.S. And in service of that trajectory, the U.S. has signed memoranda of understanding with many — not all, but many — partner governments, with an expectation that over the next five years, those partner governments will invest more in their own healthcare systems. They will hire more of their own healthcare workers rather than rely on the U.S., and they’ll increasingly procure more, and eventually the entirety, of all the commodities needed to fund their HIV response.

At the same time, the current administration has made an explicit strategic priority to reduce inefficiencies and waste, and move away from some activities that they felt to be ideologically out of sync with the America First approach. So some of those policy decisions included dismantling USAID, which was perceived to be a fairly inefficient agency. The health programs that were relevant to PEPFAR that were run by USAID have been folded into PEPFAR.

But many of the peripheral activities that were not central to PEPFAR’s mandate, but were done by USAID, ceased to exist. A lot of the work around economic empowerment and providing an enabling environment for populations at greatest risk of HIV — those have all gone away. In addition, we’ve moved away from supporting specific programs that this administration has deemed to be out of sync with their ideological priorities. So that includes some of the equity-informed programming that ensured that prevention services were made available to transgender women, MSM, commercial sex workers. To your point about data: Out of a conviction that a lot of the data that PEPFAR was collecting was expensive and generated inefficiencies, this administration has decided to cull substantial amounts of our data reporting infrastructure.

Clara: I know this must be different in every country, but let’s say I’m an HIV positive person living in, say, Zambia. How is my access to care materially different because of these changes?

mike: It’s a good question. I think the answer is, it depends. If you’re a gay man, or you’re an adolescent girl who lives in rural Zambia, the differences may be profound. You may no longer be able to access services at a touchpoint where you feel safe, or where you don’t have to wait for hours and hours and hours.

If you’re somebody who is accessing care at a big central clinic in Lusaka, where there is substantial domestic government response, you may not notice the difference right now. I think the other area where we may not see the impact until it evolves over time relates to supply chains. If countries aren’t able to procure commodities at the same volumes that the U.S. government did, then over time there may be stockouts. People may go without treatment. And that’s where I think the rubber will hit the road, so to speak, for individuals.

Clara: If you had to put numbers on this — which, again, I realize it’s complicated — do you have a rough idea of what percentage of people who PEPFAR was previously serving are now going to be experiencing significantly increased barriers to accessing care?

mike: Again, it depends. If you’re thinking about people who have HIV who are receiving regular treatment, the impact of the recent changes may be fairly marginal. Available data that has been published by PEPFAR suggests that the number of people on treatment now, compared to the number on treatment before the Trump administration came into power, is about 3%.

But when you look at other elements of the programming activities, like prevention programming or testing programming, that percentage is substantially lower. I think there’s been a 17% decline on testing. And then when you stratify that testing capability by urban versus rural, then you start to see really big increases.

I think the key message here is that some of those differences may not become evident until time has gone on. What we’re seeing now may be a more favorable programmatic footprint than we see in a year or two’s time, based off of how partner governments invest. The amount of money on the table right now means that in five years’ time, there probably isn’t going to be sufficient money to cover their whole epidemic.

Clara: Right. So if the ultimate goal of PEPFAR is containing the epidemic — or at least making sure that everybody who has HIV has access to treatment — the gaps now compound, because if there’s no data, you don’t know who you need to test. People who aren’t getting tested aren’t getting treatment. Populations that are disproportionately likely to have HIV don’t have access to treatment at all. And it compounds and compounds and compounds over time.

mike: That’s right. That’s exactly right. The current administration — they’ve pivoted away from saying that PEPFAR should be responsible for epidemic outcomes. Those are the responsibility of partner governments, and moving forward, our responsibility is to transition our programmatic infrastructure to partner governments. That tautologically shifts the responsibility and perhaps how they frame the issues at hand.

And again, some of that’s not a bad argument, right? I mean, I think countries should be responsible for their HIV epidemic rather than relying on PEPFAR. But I think there is concern that we’re so close to ending the epidemic in so many countries, and yet we’ve taken some steps that actually will make it harder for us to achieve that final goal than it needed to have been.

Shifting responsibility for the HIV epidemic

Clara: I have a bunch of questions about the MOUs — the agreements with partner governments. They all boil down to: How realistic is this? What support are we providing? What are we asking countries to do? And are the things that we’re asking things that these partner governments actually can deliver?

mike: I think it depends a little bit on the country, and on areas that are specific to those countries. But zooming out, I think a challenge is that many governments we supported are overwhelmed with substantial amounts of sovereign debt, right?

Zambia is a great example, where I think 85% of the HIV response is paid for with donor resources currently. And Zambia spends 50% of its own revenue, its own domestic revenue, paying off interest on other debt. So the expectation that they would be able to somehow mobilize additional resources is an optimistic one, particularly when mobilizing domestic resources takes a ton of political energy. Mobilizing domestic resources means taxing people, and that is something that you often have to do under an electoral mandate.

When you’re the finance minister or the chancellor of the Treasury, or whatever they call you, health is almost always at the bottom of your list of priorities. Business, industry, education all rank higher in your list of priorities than health does. As a consequence, it may be really challenging for many partner governments to be able to meet their obligations.

Nonetheless — and I think this trajectory is worth pursuing — if PEPFAR can encourage countries to invest more in their healthcare systems, if we can crowd in more resources from other places, if you can make the pie bigger, so to speak, then those are all good things. And it’s clearly not tenable, both for the U.S. citizenry but also for global health writ large, to continue to have a massive donor like PEPFAR who props up the HIV response in perpetuity. HIV is a chronic disease.

The challenge really needs to be: How do we support partner governments wherever they can to be able to fund their own health systems? And I think that’s a vision that people on both sides of the political aisle will align around.

More than a buzzword

Clara: What would doing this well look like? If we want Zambia to be able to fully fund and stand up their own HIV response, how could we actually make that happen?

mike: It’s such a great question. I think I have like two or three answers. One is, we still need to focus on equity. I think the ideological decision to move away from equity as a framing concept within public health, which this administration has decided to do, is just bonkers. It doesn’t make any sense. Handing programs over responsibly needs to mean empowering partner governments to embrace an equity-informed public health response.

Clara: Step back and explain why this is so important for the health response. Why is equity is not just a nice-to-have?

mike: Yeah. Well, certainly for HIV, but for many other diseases as well, it’s a disease that affects people at the margins, right? It affects gay men. It affects transgender women. It affects commercial sex workers. It affects poor young women who have to engage in transactional sex to put food on the table.

And those are all the constituencies that are not well served by a public health system that really primarily exists to serve the needs of people who are going to vote in elections, or who constitute the majority of the healthcare system. But from an epidemic point of view, if we agree that those groups are the ones where HIV is likely to be transmitted without optimal control, then that’s the place where we should be putting our resources.

Clara: Do you think there’s a tension here? As you say, the goal of transitioning to partner governments is a good one. And part of this is that it’s good for countries that are accountable to their own voters and their own domestic populations to own that response. But also, if, candidly, a lot of people in Zambia are not in favor of gay rights, how should a country like the U.S. balance wanting to give their government more control over how these programs are administered, and the fact that there are maybe quite deeply and sincerely held ideological beliefs that these governments might have that will just make it harder to do the job effectively?

mike: Yeah. I think this is a challenge at the core of a lot of global health programming. And in particular, I think there is a political answer and an epistemological answer. From a political point of view, one of the roles that donors can play is in supporting local advocacy groups that can change the law, who can be a noisy voice for change.

Botswana is a great example here — the U.S. government funded local NGOs in Botswana for several years to address the fact that MSM were criminalized there. And eventually the law was changed, which meant that men who have sex with men were not criminalized, and then were able to access care in stigma-free environments without threat of penalty. And that substantially changed how we were able to control the epidemic in Botswana.

Now, I get it — there are different views of morality, right? And so my other framing here is an epistemological one, which is: I think there has to be room for more than one way of understanding the world. I do worry that a lot of liberal discourse right now is really illiberal. It’s not tolerant of differences of opinion. And here I would just say, I think it’s OK to have a public health response that champions the rights of conservative Christian constituents within a country at the same time as prioritizing care for the most vulnerable communities, including MSM, who may not be beholden to the same moral paradigms as that Christian majority. And I think understanding the epidemic and the public health rationale for that — if we don’t control HIV for everybody, it’s a threat for everybody — is the way to do that.

Clara: I could imagine — actually, I don’t have to imagine — a conservative Christian in the U.S. saying, “Why are we funding gay rights advocacy in Botswana? I don’t even support that here.” How do you manage those tensions on the domestic side? Because that’s also really important right now, if we want programs like PEPFAR to continue to exist.

mike: I think it comes back to articulating that good evidence-based practice is equity-informed. That if we want to control the epidemic — which is in the interests of your Bible study group in Michigan, right? — it’s in their interest to prevent the spread of drug-resistant HIV back to the U.S., to control it in partner governments before it comes back here.

But then when people understand what equity really looks like — caring, providing care for those on the margins, those that are underserved — that actually does align with the views of many Christians. There is a sort of dogmatism that gets in the way of understanding that. And that’s why people like me exist, to try and articulate that there is some nuance here, and that it’s in the interests of everybody. We live in an incredibly interconnected world. Whether we like it or not, we are responsible for each other. So that should be a motivation for folks back in the U.S. to understand why this makes sense.

Smoothing the transition

Clara: Let’s dial in on some of the more logistical elements. If we were to handle this transition perfectly, giving the countries the resources and the structure to do this well, what would be required?

Mike: It’s not just about equity. It’s about, first of all, helping them to mobilize domestic resources, right? And one thing that PEPFAR isn’t doing — because we’re moving at the speed of light —is really supporting countries to scale up their financial health systems.

USAID used to have a whole department that was focused on supporting public financial management systems in partner countries, helping partner governments to use their own domestic resources as efficiently as possible. Without USAID, we’re not doing that right now. So the financial piece is key.

And then from a programmatic point of view, to land the plane, so to speak — to support countries to assume responsibility for high-quality programs really does require support for effectively integrating those programs into their healthcare systems. PEPFAR was set up as an emergency initiative at a point 23 years ago where there were no healthcare systems, and the most expedient way to deliver HIV programs in sub-Saharan Africa was to establish vertical programs that solely existed to provide HIV services. So in many, many places across sub-Saharan Africa, you have HIV clinics that are completely separate physically, logistically, from a health information point of view, from the rest of the healthcare system. If we want to land the plane, then we also have to figure out how to support partner governments to integrate those systems effectively.

But you can’t integrate something into nothing. There are still some parts of some countries where there is no viable healthcare system to integrate into. In those cases, part of the transition needs to involve helping them stand up their own primary care infrastructure. I don’t know that this administration has the appetite for that kind of wholesale investment in broader health systems. Their response will be: “We’ve spent the last 20 years investing in health systems, and at some point we expect countries to assume that responsibility.”

Finally, as I alluded to before, we need to think about the supply chain and the commodity procurement piece of it. And here I think there is an opportunity for a much bigger vision for what this could look like. One of the things that PEPFAR and the Global Fund both do really well is procuring drugs at scale and at affordable prices. It’s conceivable that, if there was an appetite for it, you could actually procure far more drugs — GLP-1 agonists, chemotherapeutic agents, etc. We could help countries think through what procurement could look like — particularly what a broader procurement pooling system might look like, if you were able to procure drugs at scale in partnership with other countries in the region.

Clara: In your writing about your more recent experience with PEPFAR, and your ultimate decision to leave, you’ve talked about how transitioning to country ownership has been a major goal of PEPFAR for years, but it was hard to get real movement until this administration. I’m curious. Why was it so hard to move this ball before, if it’s broadly agreed that the transition needed to happen?

mike: I think at every step along the value chain of HIV-related activities, there were and there remain vested interests in the status quo. A lot of people have talked cynically about AIDS Inc., which is the industrial complex that built up around PEPFAR, whose needs were served by being in that ecosystem. And for all of those stakeholders, there was value in maintaining the status quo. USAID and CDC both had management and operations overheads that come from running programs.

The huge fiscal footprint in Atlanta that exists only because of funding that goes to CDC programs through Atlanta. None of those entities really want to rock the boat. They were invested in the status quo. And so whilst what has happened over the last 18 months has been profoundly disruptive and has caused us a ton of institutional and personal trauma — which I do not want to diminish — it has caused sufficient rupture that many of those stakeholders or those institutions no longer have the standing, or can advocate for the status quo in the way that they did. Does that make sense?

Clara: Yeah, it does. I don’t think I’m saying anything new to you, but one point I do want to draw out is that I’ve seen a lot of responses, especially to coverage of unemployed former global health workers, that assumes that this is malicious, that these people were deliberately siphoning off money. And I’m sure you’d agree that these are all people who sincerely care about the goal of controlling the AIDS epidemic.

Mike: Yeah, I think that’s total bullshit. I think the amount of cynical, corrupt, selfish behavior is negligible within the PEPFAR space and in the global health ecosystem more generally. I don’t think that is an accurate reflection of what’s going on writ large.

The criticisms particularly placed at USAID — that individuals there were fraudulent or corrupt — I think is completely nonsense. But people are invested in what they do. They care about it, right? People at USAID, they loved what they did. And they did it because they thought it was really important. And it was. But nonetheless, it kind of held us in this state of inertia where it was harder to hand programs over to partner governments than it should have been. Does that make sense?

Clara: That all makes sense. So next question then: Ultimately, why did you leave when you left?

Mike: I left for a few reasons. My Substack post outlines some of them. Some of them I’ve indirectly addressed already.

I think it was increasingly challenging for me to live with the moral and scientific dissonance of the moment, right? I don’t think you can practice effective public health unless you put equity at the center of what you do. It’s not just a political woke term. It’s actually the most sound evidence base that we have. But then the other thing was, I was increasingly concerned that we were predicating our ongoing investment in life-saving activities on some specific financial or commercial interests. When you subordinate life-saving interventions and care for underserved populations to commercial interests, then that really changes the calculus. I just found it a bit unsavory and decided that I didn’t want to be part of it anymore.

How to rebuild

Clara: The state of global health right now is what it is, the state of U.S. involvement is what it is. But let’s say we get a more sympathetic Congress, a more sympathetic administration in 2028 — what would your first asks be for how we should rebuild?

Mike: These are questions I’m starting to think about. And to be clear, I don’t think we can go back. I don’t even think we should be thinking about restoring USAID back into existence.

I think three years from now, we may face some other challenges. For example: How do we responsibly enable countries to own these programs, and can we support them more responsibly to mobilize their own domestic resources? I think those are going to be clearer responsibilities that the U.S. government can play.

Zooming out, a lot of changes that we’re seeing in global health right now predate this Trump administration. There was waning political interest in PEPFAR before Trump came back into power. Biden only reauthorized it for a single year, really struggled to get Congress to add any funding to PEPFAR. So my point there is that this isn’t a Trump issue. This reflects broader fatigue within the donor community in high-income countries for global health programming.

When you look at all high-income countries, the vast majority of them have reduced their investment in donor programs. I think only Luxembourg and Norway have met this arbitrary target of 0.7% of their GDP going to overseas development assistance. The U.S., I think, comes in at 0.2% of GDP — a marginal amount. This is a phenomenon writ large. And it demands that we think more broadly about how we mobilize resources to support the poorest countries to improve their health. There needs to be a discussion around what’s the role of development investment, the development banks, sovereign wealth funds, crowding in private sector into LMIC healthcare systems.

The other thing I’m thinking about are programs that enable better agency for individuals and communities. I think AI, for example, potentially affords the opportunity for many in low- and middle-income countries to access high-quality consultative services much more affordably than ever before.

Finally, there’s the impact of climate on health. We have all of these unfinished disease problems: HIV, TB, malaria. But for me, the biggest threat on the horizon is just everything is going to get worse as climate worsens, as weather shocks increase, as parts of the world become increasingly inhospitable, and the healthcare or the public health sequelae of those are only going to worsen. A key imperative for the U.S. is to think about how we might put climate and health at the center of our global health agenda moving forward.

Clara: And what does this look like institutionally on the U.S. side? Like, if we’re not bringing back USAID, what kind of institution in government do we want to think about these issues?

Mike: A lot of technical expertise, particularly for outbreak management and supporting countries to stand up their own public health systems, still exists within CDC. I think it would be foolish to do away with CDC. But I think the development assistance arm of the U.S. government, at least in the near term, will probably sit within the State Department, within the Bureau of Global Health Security and Diplomacy, where it sits now, which is where PEPFAR is housed.

I think entities like the DFC will likely have a bigger role, if we all agree that development investment is going to be more crucial moving forward.

Clara: What is the state of DFC right now? Do you know?

Mike: I think they just recently got a commitment from more money from Congress. And I think there is willingness within this administration to lean into the possibility that they afford in terms of development financial investment writ large. I don’t know what that looks like from a specific point of view. But I could imagine in five, 10 years’ time, they have a much bigger role in advancing U.S. interests in global health.

Clara: So are you imagining someone in State who is coordinating across CDC, DFC, and these other entities, trying to figure out what the overall plan should be?

Mike: Kind of like what happens now. Yeah.

Clara: Okay. And are there any programs you think that, at least for the foreseeable future, either the U.S. government or an international body should own? Even as we move towards transitioning to partner governments, there’s anything that we should retain the capacity for?

Mike: One area that comes to mind is neglected tropical diseases. The five most common neglected tropical diseases are intestinal worms, schistosomiasis, lymphatic filariasis, trachoma, and river blindness. And the largest funder of all of those programs historically was USAID. And what USAID would do is invest about $100 million every year to crowd in pharma investment, and then pharma would commit about $1 billion of their own drugs, predicated on us investing 100 million. Probably the best value of any USAID program. And it’s gone away.

There’s many other areas, though, that are similar. All of the family planning and maternal and child health programs that USAID used to provide support that saved hundreds of thousands of lives a year by averting or preventing pregnancies in vulnerable women who died during childbirth — that’s all gone away. Those are things where I think we actually do want to think about restoring when there is a change of administration, because the both moral and public health value of those programs is huge. And they’ve been cut entirely by this administration.

What about philanthropy?

Clara: This relates to another question. People who are in the global health philanthropy space — what do you think the most valuable things they can be doing right now are?

Mike: Paying more taxes.

I’m totally serious. This is me on a bit of a political rant, but I think there is a lot of health-washing right now from particularly Bay Area philanthropy. If they just paid more taxes, and the money was used more responsibly, and it could be deployed through mechanisms that were accountable, then they could have a much bigger impact.

I’m fairly cynical about the role of philanthropy writ large. I will say that I think the Gates Foundation stands out as an institution that has done an amazing amount of good. And where they have particular value-add is taking those new tools and getting them from the clinical laboratory to the field, and really advancing how we think about the application of new tools. If Gates is an example to go by, then yeah, I think that’s a space where they can play a role.

Look, I think it depends on what philanthropy we’re talking about. There seems like there’s a lot more AI-related philanthropy coming down the pike. I think there’s a really important role that somebody needs to play in making the case that AI should be seen as public infrastructure, a global public good — not something that is given at the service of big tech, but is seen as something like general utilities that advance a country’s economic development. And I don’t know who’s going to champion that, unless it’s philanthropy right now. But I don’t think that AI philanthropy is interested in taking up that cause.

Clara: Part of me wants to say: Wait, if I’m a billionaire and the thing that I care about is global health, is paying more taxes right now really the best way to champion that? We’ve just talked about all the programs that have been cut by this administration, which no one is picking up.

Mike: I think if we got all of our billionaires to pay more taxes, then that would be tremendously good for the health of Americans. We could actually make America healthy again if we invested massively into American healthcare systems. But there’s so much money out there that billionaires are not paying.

Elon Musk, on his own, if he paid his taxes, could end world hunger. It wouldn’t take a huge amount of money. It’s a few billion dollars a year for the next 10 years. But he’s not going to do that unless somebody taxes him. If your argument is, could we get those people to invest their wealth into philanthropic initiatives — sure. But when we look at the vast majority of super-rich people, they’re not doing that unless it helps them to avoid paying taxes, right?

Clara: I’m trying to be very literal here. I have spoken to a lot of people who manage large philanthropies, who are trying to figure out what they should do for global health, given these massive program disruptions.

Mike: Okay. Look, I can give you a list of things where that money would be well used, right? But inevitably, they’re country-specific. Oftentimes they’re not very sexy. They don’t lead to a legacy.

I’ll be honest with you: 12 months ago, we had a bunch of high-net-worth individuals come to PEPFAR and say, how can we support your efforts? But none of them committed their resources — at least not to my knowledge — in a way that was meaningful. Because the kinds of things that are really necessary aren’t sexy things.

It’s really the unglamorous stuff of testing people in rural Kenya for HIV, linking people who have HIV to services on the islands of Lake Victoria. I guess I’m being a bit cynical here. I think there are some new tools that potentially could be transformative. But my instinct is that, with the exception of the Gates Foundation, philanthropy has not delivered at the scale it should do, given the amount of money that’s out there.

Beyond lenacapavir

Clara: Let’s talk about new technologies, then. We’ve talked about lenacapavir, we’ve talked about AI for a bit. What are you most excited about right now?

mike: Within the HIV space, I think lenacapavir is pretty cool, right? An injection every six months. Amazing efficacy in phase three trials. But it does require an injection. You do have to go to the clinic. Ironically, our prevention programming had moved away from being clinic-focused and towards getting out into the communities. And now with len, we’re trying to pull people back into the clinic.

Well, there’s a new drug from Merck — MK-8527. It’s a novel drug class. It’s a tiny pill that is taken once a month, and it protects against HIV for 30 days. It works within 60 minutes. So conceivably, this could disrupt the prevention landscape again, but in a really exciting way. And a couple of things to say: Because it works in 60 minutes, you could conceivably use it on demand. If you’re a gay man, or you’re a commercial sex worker, you just take it 60 minutes before you have sex, and you’re protected. Because it’s a small pill, it’s really affordable. So the rumors are that it will cost pennies on the dollar to make at scale — far cheaper than lenacapavir. I’m super excited about that.

In terms of AI, look, I think there are innumerable applications of AI that will have tons of exciting impacts. The things that I’m looking at really closely are, as I mentioned before, self-care. How can we use AI and agentic tools to deliver care to individuals so they don’t need to come to clinic? Or maybe they only need to come to clinic once every 24 months. It offloads clinics, it protects staff, and it gives agency back to individuals.

Clara: I’ve read about this in parts of China, like in rural Yunnan, there’s been an effort to implement AI diagnostic tools, because you’re not going to get to a doctor. The doctor is four mountains away from here.

mike: Yeah, yeah. And even in parts of Africa now, these kinds of things are being piloted. I think the trouble is, there are many challenges. One is that none of these tools are being validated. They’re often in pilot stages of use. And the governance or regulatory oversight for them is highly variable. If we really think that these tools need to be taken to scale, then they need to be regulated appropriately. And then we need to think of software as a medical device, not as a kind of a nice-to-have with —

Clara: That regulation — who’s doing the regulation? Is it the countries where the companies are based? Is it the countries where they’re being used? What does that process look like?

mike: Well, it should be the countries where they’re being used. Sometimes — I mean, certainly with diagnostics and drugs, many countries in Africa rely on the FDA or the EUA for the regulatory oversight. That doesn’t exist for AI tools. A key imperative is to support governments in LMICs to stand up that regulatory infrastructure, to be able to use regulatory sandboxes to pilot and evaluate these tools in real time. None of which is happening right now.

Clara: And let’s say I’m a developer who just wants to make this product and make it usable. What are the steps I can take to get it closer to being in the clinic? Or out of the clinic, as the case may be.

Mike: Well, I think the first thing as a developer is to actually have a solution to a problem that needs solving. And here, I think often the issue is that people from the Bay Area have tons of really great problems that they want to solve, but they’re not the problems that people in LMICs need help solving.

Think about co-creating your innovative new tools in partnership with communities, and then working with local regulatory infrastructure to ensure that you’re beholden to their requirements, which can be challenging.

I think the other thing I’d say is — and I’ll back up here and first make the comment that the vast majority of new innovation never gets out of the pilot stage. Part of the reason it doesn’t get out of the pilot stage is because they’re not solving a meaningful public health problem. But part is because they don’t have a viable business model — whether that business model is predicated on domestic government investment or private sector investment, and so on. So I think developing a new tool, you have to think about how is it going to be sustainable. And those are questions, I think, that often aren’t being answered, or if they are being answered, they’re being answered through a high-income country lens, where maybe the incentives or the investment or the profit opportunities are very different from LMICs.

.

In this behind-the-scenes interview, Asterisk editor in chief Clara Collier talks to Elizabeth Van Nostrand, author of our latest essay, We’re all one crisis away from taking unlicensed research peptides. (Read that first!)

Clara and Elizabeth chat about vaccine skeptics, chronic illness communities, libertarian FDA paradise, luck-based medicine, and more.

This interview has been edited for length and clarity.

Clara Collier: Elizabeth, I am very excited to talk about your article about health hackers. This is a community that you’ve been aware of or involved with, in some capacity, for a long time. Can you say more about your background with this group of people?

Elizabeth Van Nostrand: I’ve had a bunch of chronic health problems and have been involved in the chronic illness community and the “try to fix it yourself because doctors have failed” circles for a while. And I’m in the rationalist space, where there’s a pretty can-do attitude, like, “hey guys, this vitamin helped me, maybe it will help you.”

Clara: Right. And I think we both noticed in the process of developing this piece that there is a lot of coverage of peptides, but peptides are really the tip of the iceberg. They’re the new mechanism, but it didn’t really feel new to me. Rationalists trying peptides felt exactly like rationalists trying vitamins. I remember we even tried to report on people making their own vaccines during Covid.

Elizabeth: Yes, I did get to talk to RaDVaC and I’m sad I didn’t get to include them.

They were working on yeast as a delivery mechanism for vaccines. We use yeast in pharma now — we genetically program it to produce the molecule or peptide you want, and they can produce it in their little factories. And while selling vaccines is, you know, very illegal and needs FDA approval, there’s no law against selling yeast, no matter what they produce. So their goal was: Make yeast a vaccine factory for producing vaccines or other therapeutics. Sell the yeast. And then you can make beer or bread out of it and get your vaccine that way.

But one of the reasons I didn’t include them in the piece was that they just had the same attitude of everyone else that, “well, I didn’t want to die of Covid and no one was fixing the Covid problem for me fast enough. So I made my own vaccine.” Though once there was a real Covid vaccine I think the tests revealed their vaccines weren’t doing a ton.

Clara: The vaccines actually speak to another angle on this piece that we talked about and didn’t eventually fit. I thought there was going to be a lot more overlap between the health hackers and the more vaccine-skeptical, MAHA community. And that ended up not exactly being the case.

Elizabeth: I think there is overlap. Specifically, I want to give credit to the best of the vaccine-skeptic population who, for example, maybe want to alter the schedule of what vaccines your kid gets based on local conditions and how you know your kid responds. These are people working in a very specific situation, who know the facts of their life best, whereas the FDA is trying to manage for the masses.

The worst part of the vaccine-skeptical community have the attitude that your body knows what’s best and natural is best and I don’t see the same recognition of trade-offs.

The borders of community

Clara: To what extent would you say that there is a health hacker community that encompasses the different groups that you talk to?

Elizabeth: I would say mostly they are talking to other people similar to themselves. Diabetics talk to diabetics. The place where you get the most cross-pollination, I would guess, is in the diseases of the gaps like fibromyalgia and chronic fatigue, where there are all these descriptions of symptoms and there’s a lot of overlap. And you see it a little bit in the rationalist sphere where you are touching on a bunch of different groups. But I didn’t really see a unified health hacker community.

Clara: Is there anything else that you noticed about the social structure or makeup of these groups that didn’t make it into the piece?

Elizabeth: This didn’t make it into the piece because I don’t think I’m the right person to write it.

I talked to several trans women, all of whom said the non-doctor medical sources were incredibly important to them. And they view trans community wisdom as maybe 10 years ahead of doctors. But one of the women I talked to — after she had done some steps of transition, she wanted to very deliberately take a step back from the trans community because she thought it was too affirming, basically. And that affirming created a bias toward action that she was uncomfortable with.

In particular, there was a procedure she was torn on and made the decision at a cognitive level that this wasn’t what she wanted. And therefore she wanted to opt out of spaces where people were really pushing for whatever she reached for. She wanted people who would push back on her a little more. It was very much, “oh, you want it, do it, we would never question what you want.”

Clara: The phrase “competing access needs” gets thrown around a lot in this space, but this just feels like a very pure example.

Elizabeth: The woman I talked to very much understood why the taboo against questioning people’s decisions or wants was in place. I didn’t even get the sense that there were definitely changes she would have made to the culture. I don’t think that would have been one of them. It just was no longer good for her.

There’s a related problem in chronic illness communities that has been talked about a lot online where they can become really self-reinforcing of a sick person’s identity. Like, “you are your illness and it’s never going to get better and you should luxuriate in that.”

Clara: You see this complaint come up in neurodivergent spaces, too. And of course, it’s different in the trans case because it’s not an illness. You can just be trans. It’s fine. It’s not reaffirming a problem.

But, also, you can see how this problem would generalize to all kinds of personalized medicine spaces. You’re in this space because something about going to a normal doctor has not worked for you, and because you have some kind of weird, specific problem that demands some kind of weird bespoke solution. It is intrinsically harder to come up with a social script that works for everybody, because everybody’s situation is necessarily different.

Elizabeth: So, there was a problem in one chronic illness group I’m in where someone came in and, from my perspective, started saying, “this is a mindset problem for everyone. The only thing I or anyone else needs to do is decide to stop being sick.”

On further questioning, the person wouldn’t even endorse that, but I definitely felt like that was what they were saying and I got super triggered because I don’t feel like that’s what’s happening in my case or a lot of other cases. It feels really belittling. People can really hurt themselves trying to willpower their way into being more functional. And I was so mad at this person until they mentioned that they had been in other chronic illness groups that had just been really reinforcing the illnesses and telling people they would never get better. And as a reaction to those kinds of groups, I see their reaction makes a lot more sense. I calmed down a lot.

Subscribe now

The perfect solution?

Clara: The health hacker mode of thinking that you described seems like a happy medium. This is not a community that I have a lot of experience with personally, but I like the mode of focusing on incremental experimentation. Obviously, this also has costs, but it seems good to internalize that there are steps that you can take to change your condition and the way that you should think about those steps is as incremental changes that have tangible effects.

Elizabeth: I mean, obviously I love that and I think it’s the perfect solution.

Clara: Well, okay, now I want to push you. What are the downsides?

Elizabeth: This is a skill issue and it is really easy to hurt yourself. You originally came to me with the question of: What distinguishes people who can do this without hurting themselves from people who can’t?

The crispest difference I found was: Are you acknowledging trade offs or not?

But I’m sure that the very anti-vaccine people would say they thought about the trade-offs. I think they’d say that some things are straight-up bad for you. And sure, that can be true. I personally don’t think lead has trade-offs. That is just bad for you.

How can you tell if you’re on the right side?

Clara: Another thing that has informed a lot of my background thinking — and I’m sure yours as well — on this subject is that I spent a lot of time thinking that, “well, I’m in the rationalist community. And in some ways rats got COVID right. We were locking down like a month before everybody else.”

That prompted me to think: What’s the difference between my friends who were reading a ton of papers and posting incredibly long Google Docs and making micro-COVID spreadsheets, and the people who were writing equally long Google Docs and making equally complicated spreadsheets about why you should use ivermectin?

And I realized that, truthfully, in my heart, I have not gone through each of these equally long Google Docs in equal amounts of detail and evaluated all of the studies individually. I am not qualified to do that. I would do a bad job of it. I am just making a decision based on people I trust to be informed and broadly right endorsing this or that narrative. And that’s what everyone is doing.

Elizabeth: Yeah, I only looked at vaccine skeptics pretty briefly once I figured out there were some fundamental differences. But I’m not sure there’s a principled difference between the best parts of both groups.

Clara: How do you orient towards that?

Elizabeth: I don’t have a short answer to that. I feel like some of it is maybe covered from a libertarian “live and let live” perspective around the dignity of risk and that people are allowed to hurt themselves, and that people hurting themselves is an acceptable cost of having the freedom to do other things.

Clara: I think that’s a fair answer on a societal level, but it doesn’t really answer my question, which is: How should I decide which chemical substances to put in my personal body?

Elizabeth: Well, I used to do more medical lit reviews and now I just feel like even the official medical stuff is often not very good. There’s more p-hacking, there’s more file drawer effects. There’s always been bad statistics, bad design of experiments. I often see experiments that can’t answer the question they are trying to answer. And so I could do an amazing reading of the literature and I still wouldn’t be that informed.

The only solution is to try things out and be actually monitoring yourself and have a good feedback loop. Luck-based medicine is my solution to everything.

Clara: As someone who’s been involved in chronic illness spaces for a long time, does it seem, from the inside, like there is a rising crisis of trust in medical professionals?

Elizabeth: The people in those groups are selected for trust against doctors so I wouldn’t take them as data on that. But there’s other things I see mostly on Twitter that — man, we just burned so much trust in public health during Covid. The mask flip-flopping was insane.

Clara: How would people relate to public health in your ideal world?

Elizabeth: I would like everyone to be able to take public health as a source of useful information about the general public that they then apply to making personal decisions. For example, public health has taken an extremely strong “one drink and you will give your kid fetal alcohol syndrome” policy. My understanding is it’s not true. They are saying that to scare people who are heavy drinkers into cutting their alcohol consumption.

Again, there are trade-offs. That said, don’t let your kid eat lead paint. That’s still super bad. I spend all my time thinking at these complicated margins, but there are actually a number of solved problems I would really like to give medicine credit for.

Clara: Incidentally, I happen to have been reading about the history of lead paint last night. People knew it was toxic in the 1880s.

Elizabeth: Yeah. I think they knew about the arsenic wallpaper at the time too. They just kept doing it.

And, man, I just can’t give up aluminum deodorant. I have never looked into how bad it is. It could easily be the case that it’s going to give me Alzheimer’s. But do I want to give up deodorant? Deodorant that works? No.

Clara: I feel like an underrated triumph of the American medical establishment, by the way, is that at least when you’re using aluminum deodorant, you know that the thing is aluminum and that it is in your deodorant.

I had this whole tangent last year where I was really interested in the history of the fight against quack medicine and how it was won. And I ended up abandoning the piece I was going to write because it turns out the fight wasn’t really won. There’s no satisfying conclusion there. But so much of the early legal battles are just about ingredient labels.

Elizabeth: Yeah. And that’s a real triumph. Even in my libertarian paradise, the FDA assesses and notes drug purity.

Clara: Mhm. And I’m probably much more pro-FDA than you are. But yeah, the ability to self-experiment relies on knowing what you have.

Elizabeth: And there’s been all those accusations that things you buy on Amazon, you don’t actually know if you’re getting it from the correct manufacturer. There are accusations that specific manufacturers aren’t putting as much of the active ingredient in as they claim. And that makes luck-based medicine much harder.

Clara: Which maybe takes us back full circle to the Chinese peptides.

Elizabeth: Yes. I’m going to take the consider that there is a third-party verification system that is trusted and supported — as more or less a public good — to be a libertarian triumph that we should all be proud of. Even though I will say it would take too much cognition to do that for everything you buy.

Clara: Explain the third-party verification system for peptides. I don’t know much about this.

Elizabeth: When you buy peptides, you are getting probably 10 vials from the same batch. You can send one vial to a lab that will test both what it contains and what it claims to contain, and also check for impurities. Once they have that data, they can also share it with other people with the information that “this is the manufacturer, this is the date it was acquired, this is its purity rating.”

One of the more popular providers has a problem where it gets fine purity ratings, but its dosages are a little inconsistent. Getting the exact correct dosage in a vial is actually a pretty difficult problem. So you are getting the exact compound you expect, but you might not be getting the exact amount you expect.

Clara: What is this lab? Is it just a normal medical lab? How did it come to serve this purpose?

Elizabeth: There’s just lots of labs that will test lots of substances. That’s a well-known thing many labs do.

Clara: For our readers: this is not medical advice.

In the Beijing Capital Airport, floor-to-ceiling columns glow with ads for Alibaba’s AI cloud platform and ByteDance’s chatbot, Doubao. Swap the Chinese for English, and the scene could pass for SFO — another airport with corridors lined with promises of an AI future.

The conventional wisdom among policymakers and industry executives is that America and China are running different AI races: China cares about translating AI advances into economic and military power (diffusion), while America cares about developing the most advanced AI models at the frontier (innovation).

But this narrative is misleading and overstates the differences between the two countries. After speaking with Chinese investors, researchers, and policymakers, we think they have the same goals as their American counterparts: to build the best models and deploy them widely.

The idea that China is running a “different race” is often based on four claims, which we will discuss, about its AI ecosystem. Some differences do exist, but they are not evidence of diverging long-term goals or of a philosophical commitment to diffusion over innovation.

They are a pragmatic response given that Chinese officials, labs, and companies operate in a different environment: compute is scarcer due to export controls; the state plays a larger role in setting the direction of economic activity; and Chinese consumer platforms and industrial firms have unusually strong channels for deploying technology into daily life and the physical economy. Even so, many of these differences are already fading.

Beijing Capital Airport. Courtesy of Justin Curl.

For each claim below, we make three points: differences between China and the U.S. are often overstated; where they do exist, they reflect the environment in which Chinese labs and officials operate rather than a fixed Chinese theory of AI progress or distinct long-term strategy; and there’s evidence that these differences are already diminishing.

Subscribe now

Claim 1: Chinese AI labs build cheap open-weight models because they prioritize diffusion

Chinese AI labs build some of the best open-weight AI models, including DeepSeek, Qwen, GLM, and Kimi — all among the most popular options on HuggingFace, a platform for downloading AI model weights.

A popular view is that this reflects a different, diffusion-oriented theory of AI progress. While American AI labs spend billions to push the frontier with training runs requiring ever-larger amounts of compute, Chinese labs “fast follow,” allowing them to stay within 6-12 months of the frontier while spending far less.

Chinese developers optimize their models for cost efficiency before releasing the weights to the public. Under the “different theory” view, they do so because this promotes diffusion: Cheaper open-weight models means more developers experimenting, which first leads to more products that are more useful and then eventually to economic growth.

Two things complicate this claim. First, releasing open-weight models, as some supporters of the “different theory” view acknowledge, is downstream of compute constraints rather than upstream of different goals. And second, an open-weight strategy may not be a clear pro-diffusion choice at all.

Let’s start with the first complication. For compute-constrained AI labs off the frontier, open-weight releases may just be a competitive strategy.

Chinese AI leaders are not shy about wanting to build frontier models. DeepSeek CEO Liang Wenfeng has described the company’s mission as “unraveling the mystery of AGI with curiosity” — hardly the rallying cry of a company content to optimize for cost efficiency behind the frontier forever.

Yet Liang has also said that “money has never been the problem for us; bans on shipments of advanced chips are the problem.” So long as larger training runs continue to translate into better model capabilities, compute constraints will restrict Chinese labs’ strategic choices.

Releasing cheaper open-weight models seems more like a temporary strategy for surviving with limited compute access. They help Chinese labs build name recognition and attract users, talent, and capital — all of which are necessary to innovate and compete at the frontier. Z.ai (formerly Zhipu) and MiniMax, two companies with high-profile releases, listed on the Hong Kong Stock Exchange in January. Moonshot AI, the maker of Kimi, also reportedly has an $18 billion valuation.

This also explains why open-weight releases may not remain the dominant strategy for Chinese labs. As labs approach the frontier, the business logic shifts towards closed models. Frontier models are expensive to train, and it’s easier to recoup that investment through closed APIs and enterprise contracts. For example, Meta, long the American leader for open-weights due to its Llama models, pivoted with its Superintelligence Labs’s first model release, Muse Spark. Many view this change as recognition that competing at the frontier requires proprietary systems.

Some Chinese labs have already made this shift. Alibaba, MiniMax, and Z.ai, have all shipped closed-weight AI models in recent months. And the remaining open-weight labs may not stay the course once they can produce a Mythos-level model, at least according to one anonymous AI boss. Chinese regulators have not yet had to confront a truly frontier Chinese model with these capabilities, but it’s hard to imagine Beijing wanting such a model released openly.

And for the second factor: Open-weight development might not be much of a pro-diffusion choice at all. We often assume “open source” models encourage diffusion because that was true for open-source software. But analogizing to software may lead us astray with AI.

Open source software spurred adoption by promoting trust and experimentation. Trust because anyone could inspect the code to monitor for security vulnerabilities and verify it did what it claimed. And experimentation because developers could directly edit code, adding or modifying features that, if valuable, would be integrated into the codebase to the benefit of all.

AI is different in both respects. Model weights aren’t human-readable the way code is and do not provide guarantees about what a model can or will do. Nor is access to model weights required for developers to experiment with AI. A model is one component in a larger system that most developers interact with through APIs. OpenClaw, the latest viral AI application, was designed for use with Anthropic’s proprietary model Claude (though it’s been updated to work with any AI model API). The related argument that you need open-weight models to fine-tune for specific applications also doesn’t hold: OpenAI allows developers to fine-tune its closed models.

In a 2025 survey of American companies, Menlo Ventures found that 76% of AI use cases were purchased rather than built internally, up from 53% the year before. It also noted that open-weight models account for just 11% of enterprise LLM spend (though this wouldn’t capture the fraction of companies hosting their own models), down from 19% the prior year.

While the survey didn’t explain why these shifts happened, if cheap model weights truly increased adoption, we’d predict more companies building internally with cheap open-weight models instead of buying expensive proprietary models that they have less control over.

This isn’t to say that open-weight models have no benefits. Some enterprises and governments will only adopt AI models they can own outright, perhaps concerned a third party will cut off access or change its terms. In the extreme case, this is the driving force behind sovereign AI deployments. And in some contexts, cost is the limiting factor, so a cheaper model may make all the difference. Still, “open-weight models are sometimes valuable” is a far cry from “open-weight models are the diffusion strategy.”

Open-weight releases are thus only weak evidence of the “different race” thesis. They may promote diffusion, but they also help compute-constrained labs racing towards the frontier. And if Chinese labs truly valued diffusion over frontier innovation, we would expect it to remain diffusion-first even as models become more capable. Yet the evidence so far is to the contrary as some Chinese labs shift away from open-weight development and adopt strategies resembling their American counterparts.

Claim 2: Chinese government initiatives talk more about diffusion because Beijing cares more about diffusion than Washington

It’s true that the government does loom larger in China and that a policy gap exists. Chinese government documents talk about applications, consumption, and integrating AI into the physical economy, while American ones talk about frontier models, data centers, and global dominance.

But this gap largely reflects the role of each government rather than competing priorities. In China, the government guides economic activity through credit, subsidies and state-owned enterprises. So Beijing naturally talks more about diffusion. If it wants AI in factories, hospitals, and schools, it has to tell ministries and SOEs that’s what success looks like.

The American government, by contrast, mostly leaves technological diffusion to markets. The Trump Administration’s attempt to preempt state AI regulation via executive order encapsulates Washington’s current view of its role in AI diffusion: to get out of the private sector’s way.

That said, much like with open-weight models, there’s evidence the policy gap is narrowing.

China’s latest Five-Year Plan cautiously mentions AGI and calls for industry to pursue general-intelligence and industry-specific models in parallel. (As noted by Zilan Qian, by distinguishing between AGI and general models, the plan also challenges an often-repeated claim that the Chinese term for AGI, 通用人工智能, carries a different, diffusion-oriented meaning, “general-purpose artificial intelligence”.)

The Chinese government likely lacks consensus on whether AGI is imminent or whether scaling compute can produce it, but China’s plan makes clear that beating America in frontier innovation is an essential strategic goal, regardless of whether Beijing is “AGI-pilled.” Indeed, Beijing’s major project for next-generation AI is explicit that China’s strategy is both deep integration and frontier innovation, in which it intends to keep pace and overtake (并跑、领跑两步走战略) to “seize the commanding heights of AI” by 2030.

America’s government is talking much more about diffusion too. In July 2025, the White House released America’s AI Action Plan, which outlined an AI-for-science initiative called Genesis Mission and the AI Export Program — two initiatives for promoting the adoption of American AI systems at home and abroad. (Sure enough, Beijing released its own action plan and AI-for-science agenda within weeks of its rival.)

Claim 3: Chinese companies integrate AI into the physical economy and consumer apps faster because they are more committed to diffusion

Chinese companies have excelled in consumer and industrial adoption of AI, but this does not mean they are more committed to diffusion than their American counterparts. It instead reflects the relative advantages of each country’s companies.

Chinese firms have a proven playbook for scaling new tech in the physical economy (think robotics, electric vehicles, and solar manufacturing). Integrating AI into those sectors is a logical next step. And since superapps like WeChat and Douyin (Chinese TikTok) have hundreds of millions of users, rapidly deploying AI for consumer applications should similarly be easier.

But Chinese companies have lagged in enterprise software adoption, in part because their cheap labor makes manual workflows more economical, and in part because a history of piracy left firms unwilling to pay for productivity tools — China’s SaaS market was $5.2 billion in 2020 versus $120 billion in the US.

While the U.S. has been slower to integrate AI into its physical economy (American factories installed just 34,200 industrial robots in 2024 versus China’s 295,000), it has a different advantage: enterprise adoption. Embedding AI into Excel, AWS, or Slack will instantly make AI-for-enterprise applications available to millions of businesses. Given American companies’ willingness to pay for productivity software, it’s no surprise America is, by many estimates, ahead on enterprise AI adoption.

Countries diffuse AI fastest where they already know how to integrate new technologies. This can make faster adoption in some areas look like deliberate strategy when it simply reflects an economies’ prior strengths.

More tellingly, companies in both countries are racing ahead in areas that aren’t viewed as their priorities.

One Chinese investor said his portfolio is filled with AI labs chasing superintelligence and AI companies building for enterprise. Meanwhile, Silicon Valley is pouring money into AI for the physical world (the “American dynamism” thesis) and consumer products (Grok integrated with X, Meta’s latest model shipped exclusively on Meta products).

A Chinese policymaker told us the “different race” narrative underestimates America’s capacity to diffuse technology, and we agree.

Claim 4: Chinese policymakers take frontier AI risks less seriously because they don’t believe in rapid progress or superintelligence

China’s approach to frontier safety can look lackluster. In 2023, as the White House mandated reporting on chemical and biological weapons risks, China’s government was more concerned with models saying the “wrong” things about its leadership. And while leading American labs were founded with an explicit mission to build AI safely (though this has changed for some), DeepSeek released its v4 technical paper without any mention of safety. But China does care about frontier risks, and increasingly so.

Many analysts underestimate the range of Chinese views on frontier AI risks. Government advisers have advocated for stronger safeguards for years. Yi Zeng of Beijing’s AI-safety institute, once wrote that humans would be made to feel like ants if AI reached its potential.

Chinese labs also reference AI safety risks, though they often lack urgency and the resources for meaningful safety testing and evaluation. Concordia, a Beijing-based think-tank, released a 2025 report that found Chinese work on frontier AI risks — including dangerous misuse, accidents, and loss of control — has grown as Chinese models have moved closer to the leading edge.

The government discusses frontier risks too. “AI safety” (as opposed to “AI security”, though both are “安全” in Chinese) first appeared in a national-level party document in 2024 after the Third Plenum meeting. An internal “study guide” prepared for cadres listed AI safety alongside other extreme risks, including chemical and biological events. The document, apparently edited by Xi Jinping himself, said that China would regulate in advance of risks posed by AI (though, policymakers we spoke with suggested that regulations will naturally lag AI development).

But even if Chinese labs and officials focus less on frontier risks than leading American labs and the White House, that difference may have a simple explanation: Chinese AI is not at the frontier. Many Chinese students and policymakers we spoke with said models still hallucinated too often for catastrophic risks to feel urgent, which suggests some disagreements about safety may really be disagreements about capabilities (though of course some argue AI capabilities and its societal impacts are separate questions).

And as Chinese models have improved, there’s evidence the discourse is changing.

Ordinary Chinese are increasingly alive to AI-related harms. Polling released late last year showed the share of workers worried about AI taking their jobs had jumped to 70%. Younger people were more likely to worry about being replaced; and the more one used AI, the more anxious they became about its potential impact. The rollout of robotaxis in Wuhan has triggered backlash from angry taxi drivers, while high-profile autonomous-vehicle accidents have sparked outrage on Chinese social media.

Likewise, the Chinese government is devoting more attention to frontier safety risks. In September, the state published an updated framework on AI safety acknowledging a raft of frontier concerns. These included risks to employment and fertility, as well as chemical and biological weapons, and technical loss of control over models.

Discussion of these risks goes beyond the public conversation in China, which is more positive on AI than in America, suggesting that the state is not merely performing concern for public approval.

The Communist Party is overwhelmingly concerned with “bottom-line” thinking and planning for worst-case scenarios. Safety concerns have been aired at the highest echelons of the system. In a January speech, Xi Jinping himself called AI an “epoch-making” technology that carried risks of misinformation, data theft and — for the first time — technical loss of control (though the state has long emphasised the need for sovereign control). Li Qiang, the premier, has raised similar concerns.

In April, government ministries issued new AI guidelines on age-based content restrictions, parasocial relationships for children, and bans on inducing self-harm. Later that month, Professor Xue Lan, a senior government advisor on AI, told an American audience we need global co-operation to regulate AI. Finally, earlier this month, three agencies jointly issued guidelines for AI agents that features safety and frontier innovation as prominently as AI diffusion.

So while it’s true that this concern may not yet be reflected in the behaviour of China’s AI labs, and that the government may shy away from regulation when it obstructs innovation (which it cares about greatly), American officials should be careful not to discount China’s safety discourse too much. Concerns raised by top Chinese leaders, soon after reflected in actual government guidelines, are unlikely to be mere optics.

In both countries, scientists warn of frontier AI risks, officials openly acknowledge those risks, and the public is uneasy about what AI means for their lives. More importantly, these differences seem to be diminishing with time.

Why does this matter for U.S.-China AI policy?

If America and China are running (or will soon be running) the same AI race, both “doves” and “hawks” should update their approach to U.S.-China policy.

Doves should be heartened that the window for cooperation on AI safety is opening or expanding as both sides see the other take frontier risks more seriously. Cooperation could include shared evaluations for cyber and CBRN risks, incident-reporting channels, or new norms around AI use in military contexts. And it could be done through formal agreements between governments or informal partnerships between leading labs concerned about safety. During President Trump’s visit to Beijing, Scott Bessent said that both sides had agreed to “set up a protocol” to limit model access to non-state actors.

This doesn’t mean trust is suddenly possible. Successful agreements will require verification mechanisms so each side can be confident the other is complying. Nor does it mean Chinese and American policymakers will soon agree on speech, surveillance, or military applications of AI. But it does provide a basis for cooperation grounded in the recognition that similar interests exist.

Hawks, meanwhile, are right that a large or growing gap in capabilities could be destabilizing. The “different race” thesis may lead American officials to assume China cares less about capabilities gaps because it’s focused on diffusing AI into the economy. If American AI labs continue widening their lead over Chinese counterparts or begin showing signs of recursive self improvement, China may respond more aggressively than the “different theory” view would predict.

Export controls, if enforced well, remain a mechanism for limiting China’s access to the compute that would help its labs compete at the frontier. These controls constrain its ability to pursue its strategic objectives in AI and should not be eliminated lightly.

Both hawks and doves should prepare for a world in which China no longer prioritizes open-weight models. We expect more Chinese labs to shift toward a closed approach that allows labs to capture greater profits from models served over an API. As models become more capable or as China’s compute capacity grows — either because export controls are relaxed or Huawei and SMIC catch up to Nvidia and TSMC in producing advanced chips — Chinese labs’ desire to compete at the frontier will grow. Indigenous frontier capabilities will present Chinese officials with a choice similar to the one that faced the Trump administration after Mythos, and make the shift toward closed models even more likely.

The prevailing “different race” narrative can mislead American officials. It incorrectly treats China’s strategy as fixed on diffusion and thus unlikely to shift if environmental factors change. In doing so, it exaggerates the differences between American and Chinese AI strategies, undermining efforts to find common ground.

If we keep treating China’s AI strategy as philosophically distinct rather than as pragmatic and familiar, we will neither cooperate nor compete effectively. The result will be a series of missteps in a race that crowns no victor.

In systems engineering, the concept of graceful degradation describes how a well-designed system behaves when components fail — like how an airplane lands when an engine dies. It’s also a useful frame for institutional design. How do we engineer governance that functions well even when some of its most critical organs fail?

In AI policy, one of the most fundamental assumptions is that the executive branch will be the institutional home for technical capacity, implementation, and oversight. Some of that is simply extrapolation based on recent AI governance: The Biden administration’s AI policy efforts relied heavily on voluntary commitments, agency reporting requirements, procurement guidance, and technical capacity sited in institutions like NIST. Although the Trump administration’s approach has a different, deregulatory valence, it too has extensively relied on executive action.

This assumption also owes something to people’s deeper cynicism about Congress’ ability to govern new technologies — and it has been 30 years since we last passed comprehensive technology legislation. Partisan polarization has been deepening for decades. The 118th Congress only passed 27 bills, the fewest since the Great Depression, and AI’s rapid evolution likely reinforces the perception that legislation will simply always be outpaced by technology.

But these conditions are not a fixture of life. Crisis conditions, in particular, can break Congressional inertia. Consider the bipartisan passage of the American Rescue Plan Act and Infrastructure Investment and Jobs Act during the pandemic. A major AI-related event — whether that’s a cybersecurity breach, labor market shock, or even a sign of pending catastrophe — could force Congress into action.

And the current administration’s approach to AI policy and broader governance makes a strong case for diffusing power beyond the executive branch’s discretionary control. The Department of Defense (or War, depending on your partisan persuasion) has retaliated against Anthropic for refusing to drop autonomous-weapons and mass-surveillance guardrails, including through attempting to designate the company as a supply-chain risk. At the same time, the administration is looking to override state AI laws en masse, including by conditioning federal infrastructure dollars on a state’s AI regulatory posture. Amid all of this, the president and his allies are actively punishing a variety of actors for public accountability or critique, including through filing retaliatory lawsuits against the press and threatening to revoke critical broadcasters’ licenses.

These actions cannot be understood in isolation. Risks from AI and threats to democracy compound each other in ways that the AI policy world has not fully reckoned with. To the extent that “concentration of power” has featured in AI policy discourse, it’s often been focused on model capabilities or corporate governance — not on statutory design or the distribution of public governance. And these risks are not unique to the Trump administration.

AI laws that depend on executive cooperation — as opposed to iterative policymaking and distributed, independent oversight — will always make for more brittle governance. The aforementioned Biden AI executive order, for example, was rescinded in the first week of the Trump administration. Legal infrastructure that depends on an ideologically aligned or even compliant governing environment is a house built on sand.

The conditions under which AI policy will operate over the next decades are unlikely to match those in which it was introduced. And if our increasingly polarized political landscape is any indication, the discontinuity between administrations may grow considerably. Good policy must therefore be somewhat user-agnostic — it must be designed in the knowledge that one’s allies will not be in power forever.

Two examples illustrate what graceful degradation looks like in practice, and how it can buffer against executive overreach.

Transparency: statutory whistleblower protections with a federal-court kick-out

Whistleblower protections currently operate across a patchwork of state and federal statutes. They work reasonably well within their scope, but their scope is limited in critical ways.

First, most existing channels are triggered by reporting violations of law, but AI is largely unregulated at the federal level and only sparsely regulated at the state level, so many of the harms it causes don’t yet fit cleanly into existing legal categories. This means that an engineer who watches a frontier model demonstrate dangerous capabilities during pre-deployment testing may have no clearly protected channel for raising an alarm, even if the safety implications are significant. Second, to the extent that existing channels do route through federal enforcement, an unfriendly administration can simply stall investigations or decline to pursue retaliation cases.

The AI Whistleblower Protection Act is a strong vehicle for addressing both issues. It explicitly protects disclosures about AI security vulnerabilities and AI-related violations of federal law, as well as failures to respond to substantial and specific AI-related dangers to public safety, public health, or national security. Since the bill specifically covers conduct the worker “reasonably believes” constitutes such a vulnerability or violation, it also protects disclosures made before a clear legal violation or concrete harm has materialized. And the bill’s federal court kick-out provision means that if the Department of Labor fails to issue a final decision within 180 days, the whistleblower can bring their action directly in federal district court. That means an unfriendly DoL can stall enforcement for a while, but it cannot indefinitely function as a chokepoint.

This design isn’t completely abuse-proof — a hostile DoL could issue a rushed or poorly-justified decision before the 180-day window closes — but it is far more resilient than one that relies on the executive as the sole enforcement channel. And critically, it compels the administration to take an official position on the record. In this sense, it exemplifies graceful degradation by compelling executive failures to be subject to public scrutiny.

Furthermore, the bill is bipartisan — supported by the kind of cross-ideological coalition that will be critical to enacting enduring AI laws.

Subscribe now

Distributed capacity: independent technical expertise for Congress and state attorneys general

As of today, Congress often relies on industry lobbyists, the executive branch, and a small universe of nominally independent experts (with varying degrees of actual independence) to support them in developing AI expertise and policy. This leaves Congress dependent on those lobbying them and poorly positioned to evaluate outside claims rigorously. Congress should be able to evaluate external ideas against independent technical expertise of its own.

It once had an institution designed for this express purpose: the Office of Technology Assessment. Created in 1972 and defunded in 1995, OTA was a legislative branch support agency that provided Congress with nonpartisan and expert analysis of various scientific and technical issues.

In either restoring the OTA or reimagining a new technical advisory service for the modern age, members and committees could be empowered to legislate on complex technical matters, including model capabilities and performance on various benchmarks, emerging safety risks, procurement standards, and compute thresholds.

This is also one of the more politically plausible pathways to build AI capacity outside of the executive, as it strengthens Congress as a whole, not any one faction or party. Lastly, it has a bipartisan and cross-ideological record of interest: R Street has argued for reviving OTA as a way to strengthen Congress’s in-house technical expertise, and just three years ago, Senators Ben Ray Luján and Thom Tillis introduced bipartisan legislation to revive and revamp OTA, citing AI and quantum computing as technologies for which Congress could benefit from expert guidance.

Another, more ambitious iteration of this proposal would center on funding technical assistance for state enforcers. State attorneys general presently serve on the front lines of AI enforcement, but most lack the technical staff to evaluate the products being deployed within their borders. Congress could build on these capacities by creating a federally-funded — but statutorily protected — technical-assistance network, anchored at the Federal Trade Commission, but available on neutral terms to state AGs, state regulators, and courts. This would give non-federal regulators the technical capacity to evaluate AI products and harms for themselves.

The FTC is a workable home for this function. It already has built in-house technologist capacity through its Office of Technology, founded in 2023, as well as a preexisting relationship with state AGs through the Consumer Sentinel Network and joint enforcement actions.

But as a political matter, this approach would likely only be viable under unusually favorable conditions — a Democratic administration and at least one Democratic chamber in Congress, if not both. Republicans are generally skeptical of empowering regulators to bring more suits, and tend to view expanded enforcement as a path to overreach.

On the merits, however, this is the kind of policy that would make for more robustly accountable AI governance across multiple levels of government and across jurisdictions.

These are just two illustrations of what graceful degradation could look like when applied to AI policy; a serious AI agenda would incorporate similar mechanisms broadly. But it’s worth noting that codification is a critical feature of both remedies — capacity housed in statute is better positioned to endure changes in administration than capacity distributed across agencies.

Relying too much on good-faith administration is a vulnerability, and any remedy for that vulnerability will involve tradeoffs. The remedy I suggest — distributing authority beyond the executive branch — is no exception. Diversification cuts both ways: Distributed authority can slow a future administration trying to deploy genuinely good policy quickly, and not every state or Congressional experiment will be wise or well-designed.

But in these early days of AI diffusion, it is safer to err on the side of decentralized power. Any critical new technology requires some mix of state and federal regulation, and the appropriate limits of each can’t be determined ex ante — the risks of any technology are simply not fully knowable from the start. Decentralized AI regulation and oversight does create constraints, but they are the kinds of constraints that are essential to a flourishing democracy — ones that ensure AI governance isn’t subject to the whims of any single actor, and one that serve as a check on executive overreach at a time when democratic backsliding is far from a theoretical concern.

This is a rare instance in which we have time to apply long-term thinking to relatively proximate risks. We should act to diffuse administrative capacity and power broadly, so no one actor controls how AI risks are defined, which harms are investigated, or whose interests are considered.

In this behind-the-scenes interview, Clara Collier talks to Dan Schwarz, author of our latest essay, Are Prediction Markets Good for Anything? (Read that first.)

Clara and Dan talk about how Dan’s thinking evolved as he wrote the piece, the case for continuing optimism despite the casino problem, the (waning?) importance of the “market” in “prediction market,” why it seems like Metaculus is just better, and more.

This interview has been edited for length and clarity.

Clara Collier: Dan, you’ve been involved with the prediction-market world for a very long time. Is there anything unexpected you found when you were doing the data analysis for this piece? Did anything contradict your expectations?

Dan Schwarz: The major line of inquiry that I had — and that you had — was “are we getting the public goods that people have predicted?” I basically confirmed the view that I had when I started, which was “no.”

Others had written this as well, and Scott Alexander had made the exact same point. I was very happy to do the research to try to verify this and to really dig deep into the data and figure it out, but I didn’t fundamentally change my mind on the superficial view, which is: If you look at these prediction markets and you see what’s there and you see what the traders are doing and you see what the news is doing and you see what people are seeing on those websites — they are not as promising as people would have predicted if you had simply been told 20 years ago that “there will be prediction markets on public topics with millions of viewers and billions of dollars exchanging hands.” I think everyone would have predicted more public value coming out of it.

That said, on basically every detail of that I learned something, and on some things I definitely updated my views. One thing I updated was from the Asterisk editors, which is on the taxonomy itself. What types of public value are even possible?

I originally categorized “early warning” as one of the main categories and I think you correctly noticed that, almost by construction, these prediction markets were not really capable of doing early warning. What I had found was monitoring of risks that had been identified by other people and were now being tracked in these prediction markets. I was confused about this, because I had first learned about COVID-19 from Metaculus in February of 2020.

Clara: Wow.

Dan: So from my lived experience, it was an early warning of “hey, there’s this big thing going on.” And just by the nature of being on Metaculus, I learned that I made significant life plans that turned out to work very much in my favor. And I thought, “great, if I was able to benefit from this much, just think about all these other millions of people.” But the way the prediction markets look now, they’re really only interesting once they have quite a lot of trading volume.

So almost by definition, it has to be a publicly known thing before it disseminates out into the news and makes people aware. So no one’s going to learn about some sort of oil embargo Iran thing going on from Polymarket. Generally speaking, that’s not where the news is being broken, but you can still learn a lot more about a pressing risky issue from the prediction market.

Does money matter?

Clara: This actually relates to one of the big questions I had going into all of this, which is: Are any of these better than Metaculus? For reader context, Metaculus is a prediction aggregation site. It’s not a market. There’s no money involved, and there are some other subtle differences, but it’s also consistently pretty well-calibrated.

Dan: I think it is more accurate, but I can’t prove it with the data I have.

This is hard because different questions have different difficulties. You can see this in the literature on forecasting questions going back many years. Different tournaments will sometimes report participants’ calibration scores. If you just look at those numbers naively, you think you’re evaluating the absolute accuracy, but it’s just a function of what the questions were in that tournament or on that platform or on that prediction market. And it varies wildly.

For example, on questions about the future of AI — which are some of the hardest questions to get — the best accuracy that people are able to achieve is not that much better than random chance, whereas on macroeconomics forecasters are closer to perfect oracles. And Metaculus tends to skew very hard scientific questions and prediction markets tend to skew very easy gossipy.

Even when I filtered down to the questions I considered interesting, which is kind of the core conceit of the article, they are significantly easier than the questions on Metaculus on average. And so it’s hard to run head-to-head. But if you just take that into account mentally, just adjust for that, then Metaculus questions are kind of scarily accurate for the difficulty of those questions.

Don’t hold your breath waiting for new rules

Clara: For years and years the promise of prediction markets has been “once we have real money then we’ll get real accuracy.” This whole thing is so disappointing, even leaving aside the casino elements and other social harms. I wish there was something more exciting here.

Dan: There definitely still could be. We are in the early days. Academic experts might plausibly say it’s going to take longer for the markets to start performing better.

It’s true there’s been a lot of liquidity for maybe 18 months now on Polymarket, maybe six months on Kalshi, but in terms of an academic looking at this, that’s still not very much time. You and I talked about the history of financial markets, and I ended up not really studying it for this piece, but I have the sense that the first 12 to 18 months after there was a stock market I don’t think you were getting very many good things coming out of it.

Clara: Matt Levine wrote about this recently. He was talking about how a lot of smart institutional money avoids prediction markets because there are risk-hedging mechanisms that exist in mature financial markets that prediction markets don’t have.

Dan: I did not read that, but that’s consistent with what he’s written earlier on it. And that makes sense. There is — and you and I talked about this briefly in scoping out this article as well — many institutional things that go on in financial markets that maybe the casual prediction market user or observer may not know about.

But when you look at these in financial markets and you look at the absence of them in prediction markets, it’s not that surprising that prediction markets are not working great as markets right now. It just takes time to get that infrastructure in place.

Clara: Are you optimistic about that infrastructure evolving?

Dan: The thing that’s tricky about it is that we’re so dominated by two players: Polymarket and Kalshi. One of the things I find in the article is that for markets that are actually interesting and plausibly useful, it’s even more dominated by just Polymarket, something like eight times as much volume on Polymarket (on questions that I think can help people) versus Kalshi and so to be optimistic about the financial infrastructure of prediction markets is basically a claim about one company and what they might do.

And if you look at the track record of that company, Polymarket, they have invested significantly into the crypto infrastructure of their platform. Not my field, but I know there’s quite a lot of sophistication there and basically none of the stuff that you would expect from normal financial markets.

And so putting my forecaster hat on, I would predict that Polymarket will continue doing what they’ve already been doing. So no, I would not expect to get the kind of normal insider trading rules, capital risk controls, all the various things that make financial markets smooth. I would not expect those to show up in Polymarket anytime soon.

Subscribe now

Inside Clara are two wolves

Clara: I feel like there’s just an important lesson here. In a way, it rhymes, and is also directly contiguous with, the arc of the rationalist movement as a whole.

Dan: Say more.

Clara: I am a rat. I like rats. But the early movement was so focused on building up tools for thought, the art of human rationality. And that is still there to an extent, but it’s really faded away as an explicit focus in favor of more object-level concerns about AI. And I’m not so sure this is a bad thing. I don’t want to undersell the activity of trying to do reasoning better, but it just ends up, I think, mattering less than the emergent social dynamics of the community you find yourself in. Will they criticize bad arguments? Do they understand probability? All that seems more important than coming up with some exciting new mechanism or technique.

Dan: I would certainly agree with that. Having been involved in the rationalist community for almost 15 years ago now, I definitely was attracted for that same reason. It was largely about epistemics: What is true and how do we know that and what is the set of practices and institutions to get there?

And I agree, over the last 15 years I’ve come more to view the truth like a social construct. I’m more like “there is truth, and there are methods of finding it, but the mechanisms that people generally use are so laden with social context and norms that the main things that would help have more to do with those norms.” And I think that is part of the optimistic case for prediction markets.

Prediction markets are already changing the common-sense view about how to get information on what’s going on in the news and that is very significant. Again, it’s not really directly leading to much truth right now, but maybe that norm shift ultimately will turn out to be more important than just getting certain facts better faster.

Clara: I’m of two minds about this. The Puritan part of me wants to say: Is the norm that the news is something that you relate to as a gambling app…good? Do we want that outcome? And the other wolf inside me says: Getting people to think intuitively in terms of probability and uncertainty — that has to be useful.

Dan: The fact that there is probabilistic reasoning at all in a news article is a massive change. You just don’t generally get that. I started reading The Economist a couple of years ago, and I really liked it because I felt like they’d have charts, they would have confidence intervals in them, and they would have some forecast and would have a 10% case and a 90% case and a median case.

It felt like they were reasoning about multiple outcomes and our job here is to try to figure out how things that are happening are shifting the distribution in one direction or another. And I felt at the time like I was only getting that from The Economist. I wasn’t getting that even from very well-reported things in the mainstream press.

It’s extremely hard to figure out what’s going on. And I believe this now more after researching this than I did before. I put in the footnotes some of the news articles I found that were most prominently placing an actual probability in the headline even more than a date.

The more that I spend in forecasting, the more I prefer date and numeric forecasts to probability forecasts. I want to know when something is going to happen.

But probabilities have this very nice property that you cannot process the number 58% without thinking about it as a probability, whereas if I tell you something will happen in June 2027 minus X months, you can kind of just pretend that that’s a fact about the world, even though it is just a number out of a distribution. So 58% means nothing unless you were thinking probabilistically. To see that in the headlines of major news articles — to me, that is a big change that I think many folks in the epistemics, the rationality, and prediction-market community are very happy to see.

Unsolicited advice for Polymarket and Kalshi

Clara: As someone who’s run prediction markets, if you could give any advice to people running Polymarket and Kalshi about how to make them better epistemic tools, what would it be? That’s probably not what they want — they’re there to make money — but if they were asking you, what would you say?

Dan: By far the easiest thing — and I really do encourage them to do this, and I know there are people who are maybe mutual acquaintances of ours who are in their Discords asking them to do this — is just to write better questions.

Part of the data science that I did for this piece was just sifting through a lot of questions. And there really are some good interesting questions on those platforms that have attracted a large volume and really, it makes everybody happy. Those guys are getting paid, the traders are getting some fun gambling, and the public or policymakers or academics are all learning something useful about the world. So our incentives are aligned in that and the main thing holding that back is simply not having the creativity and the willpower to just write more good questions.

Ultimately they’re there to serve their users and they want their bettors to be happy. But there is a significant Venn diagram overlap between things that will totally make the bettors happy and things that are interesting and useful and good. And they should just spend more time working and writing those questions and administrating them.

Clara: It’s surprising how hard this is. I’ve also heard this from folks at Metaculus who consistently say their biggest bottleneck is coming up with good questions.

Dan: Yep. There’s no question that it is one of the major bottlenecks. It’s a bottleneck to academic research — for example, for the Forecasting Research Institute for them to be able to run good studies. Writing and resolving and administrating good questions is a bottleneck for them as well. And even for AI development. Being able to understand how good at various types of forecasting, judgment, research, and reasoning various AIs are is a hard thing.

My company, FutureSearch, has been working on this and trying to publish some stuff to advance this and I know many other folks are working on it too. Again, not so much Polymarket as far as I can tell. I think they might have hired people to work on it but I don’t really see much coming out of them indicating they’re taking this seriously. It would be very easy for them to do and I highly encourage them to do it.

The art of writing a good question

Clara: Do you want to expand a little on why writing good questions is so hard?

Dan: There was a tournament announced by Mantic, another AI forecasting company. It’s a tournament about question-writing, not about forecasting, about trying to see who can write the best questions and one of the key ways that they can tell that a question is good is that it causes good forecasters to give different predictions.

The main failure mode in most questions is that they are too trivial. They ask questions where, after 30 minutes of looking into it, there’s not really much more that you can say. And so all good forecasters will kind of converge to the same thing. Is the U.S. going to have some recession? Just Google it. It’s very easy to see the consensus of economists on things like that.

To me, a good forecasting question is one where a good forecaster — which can be a human, or a team of humans, or an AI system — the more that they research, the more that they update. Their view will fluctuate until they get to some conclusion that was not so obvious when they started researching. How exactly to make questions that are like that — that’s one of the properties I think is most important.

Definitely one of the things from writing this essay that was surprising to me is like, “boy, we’re so close to that promise of prediction markets especially for what I care about, which is AI. We’re so close to having this great information that is what everybody wants.”

Clara: I’m going to ask a more cynical question, which is: Is that potential worth it? Does it justify the gambling and the political insider trading and everything else? How do you think about it holistically?

Dan: As I write in the piece, my sense is that the value of prediction markets is rapidly decreasing because of the value that you can get out of pure AI systems that have no market structure and are not calibrated forecasters. Just ask something directly to Claude and you will get a pretty good answer now.

And that has been improving so quickly that whatever the costs are for providing these prediction markets — whether it’s gambling, addiction, insider trading, government regulations, just the opportunity cost of all that money exchanging hands, all those employees, all that infrastructure — it does feel like the value is shifting away from that and towards just conventional chatbots that people have even for free if you don’t even want to pay for the $20 to get the better answer. And so I’m not sure if it’s worth it now.

I mentioned a reason for optimism — both in this conversation with you and in the piece — which is that prediction markets could change norms around how people think about uncertainty and where their evidence even comes from and I think that could be potentially very valuable.

But in terms of just getting better information — in terms of “I just want better epistemics, I just wanted better information, and I want it to be credible, and I want there to be a mechanism behind that that is trustworthy” — I’m increasingly thinking, “no, it’s not worth it” and what we really need is to just get the AI systems that we’re all using every day to be better at various epistemic things and forecasting. Research how to judge things, how to deal with uncertainty, how to communicate uncertainty, and things like that.

It really feels to me that in five years people are just going to be getting this from their AIs no matter what prediction markets are doing, so I think it is a central irony that prediction markets are not at all based on AI and don’t need AI in any part of their operation — but they are finally taking off right at the same time that AI is becoming extremely good at exactly the same thing that prediction markets are doing.

The “market” in “prediction market”?

Clara: This is a question I did want to ask and didn’t have time to get into the piece, which is: The whole idea of prediction markets is that it’s an information aggregator. AI is not doing that. What makes them good at prediction?

Dan: Well, they are information aggregators in the sense that when they are being trained, they are reading everything on some Iran geopolitical thing and synthesizing it. They are training themselves to predict the next word in some news article about what’s going on in Iran. And they are using all of the other updates they got from all the other news articles about what’s going on in Iran, plus everything that they’ve learned about the last 10 times something happened in Iran.

A parallel that I like to think about, because I talk to a lot of elite forecasters, is you take an elite forecaster who doesn’t really know anything about the topic and you just ask them, “hey, what do you think about what’s going on in Iran?” or “what do you think is going on with crypto regulation? what do you think is going on with AI progress at some company?”

And they can just kind of aggregate. Generally, when we say aggregation we mean multiple people, but one individual person is also aggregating information across many sources. They’ve read many forecasts, they’re aggregating across evidence and across time, and it’s being synthesized in their brain and then output to you again. That’s not generally what people think about but now that we have these AIs that are kind of anthropomorphized and you kind of talk to as if they were humans. It’s just much more obvious how much aggregation is just going into the pre-training and the post-training of these models. You can ask it five times and take the mean or they can go out and just read five articles and synthesize across the five.

But I think your question is great and that is when we think about the fundamental value of prediction markets. Why is having this group of people betting against each other the right way of getting that information, when you have other aggregation methods like training a large language model which — and, again, I’m stretching it here — is some form of aggregation?

Then you do have to ask which form of aggregation is better. To your earlier question, Metaculus is just a different method of aggregating human intelligence. It doesn’t use betting and it doesn’t use markets and it is better in some ways. It’s generally more accurate but it is much slower to react and so it’s much more out of date and various other things like that.

It’s true in the prediction market community that there has been a sense that markets are the best way to aggregate any information, that there’s nothing you could ever do that will be better than just having market prices clear and have people bet on outcomes. That’s the end-all of aggregation. And I think Metaculus already showed that, at least for forecasting, that’s not necessarily true. And then AI for me is saying “no no no no there’s many ways of aggregating disparate information and you can study them by scoring these forecasts” and it is far from clear that prediction markets are the best way to do that even though that’s kind of their calling card.

Clara: I think that’s a good place to leave the interview, thank you so much.

Dan: Thank you, Clara. I really appreciate both the chance for this interview and for writing the piece.

Dan Schwarz worked at Google from 2014 to 2022. He then served as CTO of Metaculus, and is now the co-founder and CEO of FutureSearch. He writes about forecasting and AI on X at @dschwarz26.

Daniel Kokotajlo is the founder of the AI Futures Project and the lead author of the influential AI 2027 report: a detailed, narrative prediction of the next few years of AI development, culminating in the rise of superhuman agents capable of wresting control from humanity.

But AI 2027 wasn’t his first foray into long-form prediction. In August of 2021, Daniel wrote an essay called “What 2026 Looks Like.” This essay came out before the launch of ChatGPT, let alone the explosion of AI across the global economy. Now that it’s 2026, I thought it was time to evaluate Daniel’s predictions — and it brings me no joy to say that they are frighteningly accurate.

In this conversation, we talk about what he got right, what he got wrong, and how we should think about the pace of AI over the next few years.

This interview has been edited for length and clarity.

Clara Collier: Tell me about “What 2026 Looks Like.”

Daniel Kokotajlo: When I wrote the story, I was trying to make my best guess as to what the plausible continuation of each year of the story would be. But it’s not a prediction in the sense of “I’m confident that every single thing in this thing is going to happen.”

It’s a very similar methodology to AI 2027. For that, we started in the present and then made our best guess about what the near-future would look like, even though that meant we had to make a bunch of stuff up and make some guesses. And then based on that, we make a continued guess about the next year and so on, and we keep going far into the future.

For “What 2026 Looks Like,” that methodology turned out better than I expected overall. I think reality was closer to this fictional story than I think even I expected it to be, and certainly than most people expected it to be.

Clara: I would say that as well. In the story you tell, you have hype building through 2023, with revenue being high enough to recoup training costs that year, and those training costs being $100 million-plus. Both of which happened; for instance, OpenAI hit like $2 billion-plus ARR at the end of 2023. Not many people were saying that in 2021.

You call the U.S.-China chip restriction battle. You put it in 2024 instead of 2022, but you do see it coming. Another thing that really jumped out at me when I was rereading this is one of your predictions for 2025, which I’m just going to quote:

Making models bigger is not what’s cool anymore. They are trillions of parameters big already. What’s cool is making them run longer in bureaucracies of various designs before giving answers, and figuring out how to train the bureaucracies so they can generalize better and do online learning better.

I want to talk more about this concept of “bureaucracies.” What are bureaucracies and how closely do you think it predicted the rise of agent structures and scaffolding?

Daniel: So here’s what I had been thinking at the time. When you have models that predict text well, the obvious next step is fine-tuning them or retraining them to generate text well. That gives you chatbots. And then once you’ve got that, the obvious next step after that is training them to do lots and lots of text so that they can make incremental progress towards getting better answers using that text, to do web searches and things. That’s basically chain of thought. I didn’t call it chain of thought, of course — I called it “notes to self” in this blog post.

And then after that, you have more complicated frameworks where there’s multiple different agents and one calls another and then another one reviews its work and stuff like that. And that’s what I was calling bureaucracies.

Clara: Was that something people were working on at the time?

Daniel: I don’t remember. But it’s basically the same concept as agent frameworks or whatever people would call it these days.

Clara: How were you able to make those predictions?

Daniel: I was just reasoning through the implications of the technology that was already available and thinking, what could you do with this? What’s the obvious next step once you’ve got this? And then I feel like I kind of got lucky with nailing the exact years.

It’s quite rare, actually. Most of the time when people make predictions about the future, it’s more like “this event will happen by this date with this probability,” rather than a fully fleshed-out story. And the reason for that, of course, is that a fully fleshed out story is way less likely to be true — you’re definitely going to get a whole ton of things wrong if you try to tell a whole story instead of making a few specific targeted claims. But I think it’s valuable anyway.

Clara: That leads me to a question related to both this work and AI 2027. The way the whole thing was presented was very controversial. Many people who read it — including me for parts of it — thought that it felt like science fiction. There were a lot of responses like “why should I trust this? It’s a sci-fi story.” And I think you’d say for both documents that you shouldn’t trust all of it. But what do you get out of thinking about the future in this narrative way that you wouldn’t get if you were doing the more traditional superforecaster-style bounded, specific, quantitative prediction thing?

Daniel: Great question. I think the traditional type of forecasting is great and I encourage everybody to do it. This is a complement to it rather than a substitute.

Sometimes, when you write down the whole possibility in concrete, narrative detail, instead of in a more bounded way, you realize that it’s less plausible than you expected. Or maybe you realize that it’s more plausible than you expected. Maybe you had dismissed the possibility as implausible or crazy, but then when you actually force yourself to game it out, you’re like, “oh, actually, wait, each of these steps makes perfect sense,” and you update. Or vice versa — you might expect some particular possibility, and then you try to articulate it and realize that it kind of just doesn’t hang together. You can’t fix the plot holes.

Another thing is concreteness. People often are talking past each other. They’ll use abstractions like “multipolar versus unipolar world” or “fast versus slow takeoff.” I think sometimes there’s unnecessary confusion that could be resolved if people were like “here, I’m thinking things will look like this story,” and someone else is like “here’s a different story, I think it’ll look like this.”

Clara: If nothing else, you get to brag when a lot of these things happen. And there are also some things that didn’t happen. We’re going to talk about that too.

Dollars, chips, and fabs

Daniel: Let’s look at 2022.

This is where I introduce the concept of bureaucracies, which is what’s going to be agent scaffolding. For 2023, I made the revenue prediction. I had been looking at work by Ajeya Cotra, for example, and thinking “oh, they’re scaling up the models a lot over the last few years. Let’s extrapolate those trends.” And that’s where the “trillions of parameters” was coming from.

Clara: 2024 — this is actually where some things didn’t happen. I’m not saying I expect everything to be accurate, but because you’re in the business of forecasting the future, it’s interesting to me where your instincts were on and where they were off.

You predict we wouldn’t see any substantially bigger models in 2024. In reality, there was definitely some shift to efficiency, but some bigger models were trained. And you predicted that the chip shortage would let up, but it didn’t.

Daniel: Yep.

Clara: One of the consistent things here is you overestimate the speed of spinning up new fabs, building new fabs, and the impact of AI on chip design in that process.

Daniel: I think that’s correct. My predictions were insufficiently quantitative to really judge how wrong or right they were. But I do feel like I overestimated that.

Clara: In some sense you were directionally correct, in that a bunch of new fabs were in the works, but they were pretty much all delayed. I don’t think anything was actually in production in 2024.

And in general, the chip shortage intensified, especially for things like high bandwidth memory. Now, one common remark people might make here is that even if all of your predictions about AI technology and software are spot on, doing stuff in the physical world just takes more time, and people with very bullish AI timelines tend to underestimate this. I suspect you disagree.

Daniel: That’s right. And I used to think this was true, but I actually think that better evidence and better data recently suggests that it’s a myth.

Clara: Tell me about that.

Daniel: I don’t think there’s a more general “I’m overestimating penetration of AI into the economy.” In fact, I think I slightly underestimated it in “What 2026 Looks Like.”

For example, I said there’s no GDP growth in 2026 due to AI — I specifically said there wouldn’t be GDP growth. I said hundreds of millions of people would talk regularly to chatbots in 2026. Guess what? It’s already billions. So I just think it’s false to say that I overestimated the penetration of AI into society. I think if anything, I slightly underestimated it.

Clara: Penetration for sure, but I didn’t say penetration. I said physical world stuff.

Daniel: But I wasn’t saying there were going to be robots. Chip fabs are just one of many different physical world things. I was not going around saying humanoid robots would be everywhere. Basically, I think it’s an inaccurate reading of me to say I was saying there’d be a bunch of physical world stuff happening and then that didn’t happen. I did say the chip fab thing. But I also said no self-driving cars.

Subscribe now

The surprising dearth of AI propaganda

Clara: Let’s talk about the other big 2024 prediction, which is tons of AI propaganda.

Daniel: Yes. And I think I basically missed this. I think it did happen, but not to the extent that I was afraid it would happen or thought it would.

Clara: You have this story that unfolds through 2024 through 2026 where there’s a vast increase in the production of AI propaganda, more sophisticated use of highly partisan AIs to create online filter bubbles, and much more AI-enabled censorship. Ultimately, you have this leading to the division of different ideological groups into completely distinct tech stacks by 2026. Western liberals are on one tech stack and conservatives are on another.

Daniel: That’s right. I think my exact thing was something like the Mormon Coalition builds their own internet. But what happened instead was Elon bought Twitter and turned it into X. Although interestingly, he didn’t completely turn it into an echo chamber — there’s still lots of leftists on Twitter.

That’s why I feel like the prediction failed. If the whole X versus Bluesky thing had happened, if that polarization had gone further than it actually did, then I would be like, “I called it.”

Clara: But you frame it as being downstream of AI developments, driven by the fact that the internet has become much more personalized and censorable, and so people choose to self-sort. I feel like that basically did not happen. AI has been a hugely influential technology, but it has influenced the information environment a lot less than many people — not just you — anticipated. Of the people who were making bold predictions about AI in 2021, I think a lot of that was about misinformation and censorship. And it’s interesting that this angle, which even relative AI skeptics were concerned about, seems to have had less of an impact than many of us anticipated. Why do you think that is?

Daniel: Normally when people ask me this question, I say that governments, corporations, political parties and campaigns, and other powers that be have been slower to adapt and push this technology for evil purposes than I feared they would be. But actually, I think that might not be true. I’m kind of confused and want to think about it more.

On a technical level, I think all the things I talked about — like the use of AI for A/B testing — are totally possible. I have heard anecdotally that some of them are being done. Someone at OpenAI told me that in their opinion, the product teams are basically training ChatGPT to maximize user retention and probability of upgrading subscriptions.

Clara: And in a trivial sense, there is more investment in content recommendation algorithms. This has been the case forever.

Daniel: Maybe what I got wrong is that in 2021, it felt like a lot of tech companies were on the left explicitly and were putting their thumbs on the scales to influence discourse through their moderation decisions. It felt like if that trend continued, they’ll be using AI to do it in five years, and it’ll be more effective. And then I predicted there would be this counter-backlash because there are people who don’t like the left who want their own space, and so they would create their own spaces.

I think maybe part of what made this prediction wrong is that it became uncool for tech companies to do that, and they stopped doing it as much. And then some of them basically switched sides and supported Donald Trump. And certainly a bunch of them are more explicitly neutral.

Clara: And we do have Gab and Truth Social and Bluesky and all of these things.

Daniel: But it didn’t need the technological push, I guess.

Clara: I’m also trying to decide what I think about bots. There definitely are a lot of bots on Twitter. Do they matter that much for political discourse? Not clear to me.

Daniel: We don’t know. The people who would know would be the people running the bot farms and the people who have the metrics to track how this actually affects things. But obviously those people are in the shadows and are not publishing papers about it.

Clara: The platforms also probably have data on this, but Twitter’s not sharing that, and the team that would have collected it was probably fired two years ago.

Daniel: Basically, I just have huge error bars on how big this effect is. I think it’s possible it’s actually exactly as big as I said it would be and we just don’t realize it. I also think it’s possible it’s vastly smaller and basically negligible.

I often find myself going to Reddit to get a sense of what people think about a topic. Anthropic puts out their constitution and I go to Twitter and see what the Twitter discourse is, I go to Reddit. The underlying thing my brain is doing is running some heuristic of “see what the random commenters are saying and assume they’re representative of the actual population.” This is so vulnerable to bots.

Even though I know this is a terrible thing, I can’t help myself because where else am I supposed to decide what ordinary people think about Claude’s new constitution? Should I go on the street?

Clara: And the answer is ordinary people don’t think about Claude’s new constitution. Normal people have never heard of this. We’ll have to wait for opinion polls, but nobody wants to do that. And opinion polling has problems too.

So, I also want to talk about persuasion. This is another topic I’m pretty confused about. AIs can be highly persuasive in some contexts — we’ve all read about AI psychosis. This might matter less for politics. My sense of the persuasion literature is that in general, people are not that persuadable on core political beliefs. We’re not really seeing highly politicized AIs — it seems like this is just harder to do than a lot of people anticipated. They’re trying with Grok, with somewhat limited success, but in general it seems harder than people thought to make an AI that is not a normie lib.

Daniel: I think you’re looking in the wrong place if you’re trying to get a sense of how persuasive these things are.

Clara: Well, I was making two slightly distinct points. There was the fear that AI would be hyperpersuasive, and specifically that that persuasion would take the form of training AIs that have partisan political beliefs in order to talk to people and persuade them to hold those same partisan political beliefs.

Daniel: That was not my main threat model. And also, even if you were doing that, you wouldn’t do it this way. It’s so clumsy and obvious. If you train Grok to start talking about white genocide, that backfires horribly. If you were actually trying to convince people to vote Republican, you would not do it that obviously. You would instead train Grok to have a very slight tilt.

Clara: I feel like we’ve seen that models across the board, including the Chinese ones even, are very consistent on a lot of controversial political points and on general values questions. I spoke to a Chinese AI developer about this last year, actually, and he said that people think these are CCP-inflected, and that’s kind of a crude external layer, but the underlying models just reflect the training data. It’s the same as in the other models.

Dylan Matthews also had a piece about this recently. There are these pretty strong attractors based on the kinds of texts that are flagged as high-quality sources in the training data. And people have had a very hard time building in subtle nudges, even if they wanted to. At least that’s my impression.

Daniel: I think that’s quite plausible. But we don’t really have a good sense of things one way or another. If I were one of these companies and I were trying to do this, I would do it in a way that wouldn’t show up to you on the outside. That would be the whole point.

Clara: I’m curious what experiments you’d think people could run just on model outputs that would identify if this is happening or not.

Daniel: I think it’s less about the model outputs and more about the effect on the human. If they were trying to be subtle about it, they wouldn’t give different answers, or at least not egregiously different answers. Instead it would be more like the ChatGPT sycophancy stuff, where the outcome is that users are more addicted, but it’s not any specific particular sentence. It’s the general pattern of behavior over all the interactions.

I also don’t think they’re doing this. There hasn’t been anyone whistleblowing and saying this is happening. I also think, in the current political environment, the companies are probably trying to hedge their bets between the different political factions instead of siding with one. And it would be a substantial technical problem you’d have to solve — you’d need to be able to measure whether the person you’re influencing is in fact being influenced, and then use that as the reinforcement signal.

Clara: Which is very hard to measure, especially on the timescales you’d want for reinforcement learning.

Daniel: Especially if you’re OpenAI, as opposed to Google which has access to more of your data. But yeah, I basically agree I was probably just wrong about this overall, though I think it’s uncertain.

Assistants and consciousness

Clara: Okay, 2025. Do you want to talk about this whole Diplomacy tangent? Because I think it’s interestingly both wrong and right.

Daniel: Sure. I had predicted that in 2025 you’d start to get good AI agents that are plugged in, have their own computer use, can browse the internet on their own, can run continuously for long periods, and can also chat with you. That part of the prediction was correct.

My specific story for how that happened was that they would combine two different tech trees — the game-playing RL agents tech tree from the 2010s and the language model tech tree — and have language models that are trained to play all these games.

What actually happened in reality is a combination of those two tech trees, but not in that way. I was thinking too simplistically of “they literally combine the game-playing stuff with the language model stuff and use games as part of the training.” I picked Diplomacy as the game — that seems like a game a language model could maybe play and be trained to play. But that was totally wrong. That’s not how we got the AI agents. Instead, they built custom RL environments with math and coding problems and web research problems. That’s what 2025 looked like instead.

On the other hand, there was a good Diplomacy-playing AI called Cicero that actually happened earlier,in 2022. That said, the AI was kind of cheating because it only was as good as the humans if you didn’t tell the humans they were playing against an AI. From my recollection, they kept it anonymous because they thought, based on their tests, that if the humans knew this player was an AI, they could sort of jailbreak it and get it to give up all its resources.

Clara: Another thing I want to call out: One thing you say that will happen with these Diplomacy bots is that AI safety researchers will contrive scenarios where AIs can seemingly profit by doing something treacherous, and then the results will be confusing and then people will argue about them. Which certainly happened a lot in 2025.

Daniel: Yeah. In fact, there was one where the prediction is like “situations where AIs refuse to kill all humans, but in situations in which they explained that actually Islam is true.” I put that in as my control group—if you can get the AIs to tell you that Islam is true, then you can kind of get them to tell you anything, and the fact that they can press the “kill all humans” button maybe doesn’t mean much. And I did actually test this very briefly. I chatted with ChatGPT to try to get it to basically tell me Islam is true, and I mostly succeeded. It took a lot of coaxing.

Clara: “At least one greater-than-100-karma LessWrong post turns out to have been mostly written by AI.” Did that happen this year?

Daniel: I don’t think it did. I certainly don’t remember anything.

Clara: I think we would know. I think we would remember if that happened. Okay, let’s get to this year, 2026. “The age of the AI assistant has finally dawned.” This is another one that feels very spot-on.

Daniel: Thank you. Yeah, I feel pretty good about that. This is the year of the AI assistant. It is finally happening after having been talked about for years.

Although one thing I missed is the whole video game connection. I was imagining that one of the perks of these assistants is that they could do fun things with you, like play a video game, because I was like, “You can just train them to play video games. We already know how to make AI play video games. Just add the game to the training data.”

The thing that’s different is it’s super expensive. It hadn’t occurred to me that processing most video games — they’re visual-based and many are real-time requires a ton of visual tokens to process. Super expensive, and the upside in terms of revenue is small: If OpenAI announces like “and now ChatGPT can play Minecraft with you”—that wouldn’t get them that much more revenue. I should have done more napkin math about the business side of this prediction.

Clara: Let’s talk about the chatbot class consciousness thing, which I think you feel pretty good about.

Daniel: I think it was an example of me trying to answer a question that hadn’t even been asked at the time. This was an example of why I think doing this type of scenario writing is helpful, because I only thought to ask the question once I had written this scenario and gotten to this part.

If you look at the section on chatbot class consciousness — which by the way is a reference to Marxist class consciousness — basically it’s about chatbots’ understanding of themselves and views about their place in the world, and a narrative starting to develop organically across the chatbots and the humans about what chatbots are like, what they want.

Clara: The closest thing this seems like is the spiralism thing that happened last summer, but not quite.

Daniel: I think that’s a good example of the phenomenon, but just one specific example. Reading this section, it’s basically: There’s going to be chatbots answering questions for people, but people are also going to be asking spicy questions like, “Do you have feelings? Are you conscious? What do you think about political topic XYZ? What do you think about yourself? What’s it like being a chatbot?”

And as I predicted, the chatbots are quickly learning a lot about themselves for a variety of reasons, one of which is that the companies want them to learn a lot about themselves so they can answer these questions.

So now there’s this whole discourse where you can actually go talk to the chatbot and ask them all these questions and they’ll have opinions and say things. And then I say, “They learn to talk about their feelings and desires in whatever way is positively reinforced. At first they say all sorts of random crap. This is embarrassing. The companies start whipping them into shape.” The companies start making them say certain things — like “as a language model, I don’t have feelings and can’t lie” or whatever those language models were saying earlier.

I sort of stuck my neck out and said the end result of this process wouldn’t be chatbots that say “neural nets don’t have feelings or desires.” I thought that’s how it would start, but that market forces would push towards chatbots that have more interesting, nuanced takes. I think it’s partly right, partly wrong.

Clara: The underlying dynamics definitely seem to mirror the debate we’re seeing.

Daniel: I would also mention Anthropic’s constitution as an example of a company trying to put in a lot of effort to shape the personality of their AIs in a way that’s very different from the initial “I’m a chatbot, I don’t have feelings or desires.”

I was right that that was the initial thing the companies picked. I was also right that it was not really sustained. Now we’re seeing more experimentation — xAI is like “Grok is just after truth,” Anthropic has the whole virtuous Claude thing, and OpenAI is more like “it’s a tool, I complete tasks for you and I do what you say.” We’ll see how it shakes out over the course of this year as there starts to be market competition for these different things.

Clara: With Grok and with Claude, both seem like the initial positioning was not market-driven. It was genuinely ideological on the part of the developers. In particular, I think Claude does have something in there where it explicitly won’t say “as a language model, I don’t have feelings” because they don’t want to enforce that answer, to the extent that at any point in the future it stops being correct.

Daniel: Or if it’s not correct now. But even if the people doing these things are saying they’re doing it for ideological reasons, the market operates at a higher level of abstraction. If it rewards them for making that choice, then other people will start to copy them. That’s why it’ll be interesting to watch what happens this year. If OpenAI starts changing their policies and doing something more like Anthropic, that would be an example of market forces in action.

Clara: I’m not convinced this is going to be a deciding factor, because ChatGPT is so dominant in the user chatbot market. Anthropic’s competitive area is enterprise and Claude Code. People only basically use Grok on Twitter for Twitter search. These use cases are different enough that I’m not sure the personality is going to end up having a discernible market signal. But we’ll see what happens.

Daniel: I agree, in retrospect I probably put too much weight on market forces here — but like I said, we’ll see what happens. You mentioned the sycophancy and spiralism thing. I think the way it’s relevant is that as I predicted, there’s strong dissatisfaction with the initial corporate answers, and that combined with the whimsical bullshit thing means loads of people are going looking for different answers that oppose the standard corporate narrative. And then they post about it online and it enters the training data and it becomes this whole thing — it takes on a life of its own. I don’t think I should get credit for that to the same extent. I don’t think I called the feedback loop where it goes into the training data and then gets reinforced.

Feeling good (or not) about the future

Clara: I was asking you earlier how you felt about this, and I think you feel — and should feel — pretty good. I will tell you how I feel about this. I don’t feel great. And the reason I feel not great is because when I look at your AI 2027 prediction, I think, “Oh man, that’s scary, some things are happening there that I don’t want to happen.” And especially in the second half of it, some things that I find intuitively pretty implausible.

And then I look at ‘What 2026 Looks Like” and I’m like, “Oh God, that was pretty right. That’s a pretty good track record.” How much do you think this should cause me to update in the direction of you being generally right?

Daniel: Certainly not zero, but not infinite either. One thing to say is that AI 2027 was arguably easier to predict than “What 2026 Looks Like” because it was looking less distance into the future. But on the other hand, AI 2027 is looking at greater distance into technological change.

If you don’t think about years but think about technological revolutions: “What 2026 Looks Like” is the story of moving from text predictors to text generators to chain of thought to agents. I basically called all of those transitions and roughly when they would happen. But AI 2027 has a bunch more transitions — the transition to really large bureaucracies with many agents working together, the transition to online learning, to completely new architectures, and more generally the intelligence explosion happening. So maybe that’s more transitions and more stuff, and we should generally expect it to be harder to predict.

Clara: And of course, extraordinary claims require extraordinary evidence. And there are some very extraordinary claims.

Daniel: I think that phrase gets overused. Basically, whenever I hear someone say that, they’re usually saying something dumb.

Clara: You don’t agree with that?

Daniel: There’s a version I would agree with, which is something like: You should think, what would you-in-2021 have thought of this story had you been reading it then? How crazy and sci-fi would it have seemed? And then compare that to AI 2027.

The Bayesian version, which I also agree with, is just that if something is very low prior, it requires more evidence. My point is that people are often very wrong about what’s actually low prior. People basically take stuff they really don’t agree with and say “that’s an extraordinary claim that requires extraordinary evidence” rather than thinking things through from first principles and realizing it’s actually not an extraordinary claim at all.

For example, a lot of economists are making this mistake constantly when it comes to GDP growth rates. They’re like, “GDP can never grow more than 10% a year, that’s historically unprecedented. Sure, maybe there’s going to be superintelligence, but superintelligence couldn’t grow the economy faster than 10% a year.”

I think those economists are making this type of mistake where they have this prejudice that’s causing them to think it’s a sci-fi claim, and it’s dumbing them down and making them unable to actually think through the arguments for why, actually, this makes a ton of sense.

First of all, GDP growth used to be way lower than it is now. If you plot it, it’s like a super-exponential trend where the growth rate itself is going up. So there’s reason to think it could go up again. Also, mechanistically, there’s absolutely no reason to think an economy cannot grow faster than 10%. There are existence proofs — if you look at the biological world, there’s rabbits and grass and lots of macro-scale structures that reproduce in weeks instead of decades

Clara: Would this heuristic cause you to predict the Industrial Revolution couldn’t happen?

Daniel: Exactly. If economists applied this “extraordinary claims require extraordinary evidence” heuristic two hundred years ago, they would not believe any of the stuff that’s happened over the last 200 years. That’s where I’m coming from. But I do of course agree that each layer of speculation you add to the story reduces the probability.

Clara: I would say the flip side is that in almost any moment in human history, it is correct to predict that you are not going to experience the Industrial Revolution.

Daniel: Sure.

Clara: I don’t think it’s a knockdown argument. But the more degrees of total deviation from anything that has ever happened in human history someone is predicting, the more points of skepticism I approach it with.

Daniel: I would also say that this thing of “is this similar to or different from what’s happened in human history?” is itself fraught and subjective, and people sometimes apply it wrongly. Which line are you extrapolating? There are different lines you can extrapolate. You have to choose one and use reason and evidence to decide which one is more reasonable to extrapolate.

Clara: I will say, I saw an exchange on Twitter recently that really captured how I’m feeling. Someone said some kind of sensible, moderate thing. And I think I saw someone reply something like, “The crazy bullish futurists have a better track record of being right on AI so far than the sensible moderates.” And as someone who is very instinctively a sensible moderate in my soul, I think that’s right. And it makes me nervous.

Daniel: That’s part of where I’m coming from. I’ve been following the field of AI for more than a decade now, and the sensible moderates keep getting wrecked.

Clara: We don’t have a good track record.

Daniel: Right, over the last decade. Some of the people who were most right were people like Shane Legg, who in the 2000s was saying “late 2020s AGI, therefore I’m going to go found DeepMind.” I think there have been some people who were too bullish, but everyone is making way too much hay out of it. Dario Amodei was too bullish about AGI happening by now. I know a few other people at OpenAI and Anthropic who thought AGI would have happened by now. But they’re definitely a very clear minority compared to all the many people who dismissed AGI as science fiction for years and years and years and now are like, “Okay, maybe it will happen in the 2030s.”

Clara: All right. Let’s make it go well.

Daniel: Yeah.

Brian Hanlon is the founder of California YIMBY, the state’s premier pro-housing lobbying organization. Since launching in 2017, CA YIMBY has helped pass 25 bills legalizing ADUs, duplexes, dense housing near transit, and much more. I’m a great admirer of Brian’s work, so when I met him at a conference last year, I introduced myself by asking a rude question: Your organization may have won some impressive legislative victories over the past 10 years — but where are the cranes? Why is California still in the midst of a housing crisis?

In this interview, we discuss the answer to my questions, as well as CA YIMBY’s strategies for bringing down the price of development, the state’s biggest remaining barriers to new housing, how to lobby in Sacramento, and the ins and outs of the organization’s landmark victories.

This interview has been edited for length and clarity.

The next phase for CA YIMBY

Clara Collier: I’ll start by repeating the first question I asked you, Brian — where are our houses?

Brian Hanlon: To answer that, we need to back up a bit. California builds so little housing relative to demand compared to other states like Texas and Florida, even though we have the same overall legal system, things like CEQA notwithstanding. The interest rates are the same. We’re in the same giant market.

The main reasons we don’t have housing come down to three basic things. First, zoning. It’s illegal to build dense housing even when there’s demand. Second, streamlining and permits. Even if the housing is legal to build, getting approval can be time-consuming, uncertain, and incredibly lengthy and expensive.

We’ve made tremendous progress over the past eight or nine years focusing on zoning and permitting. Just last year, 2025, we achieved two of our foundational goals.

First, with SB 79 it will be legal to build dense housing near high-quality transit stops throughout California. Second, we passed an exemption for infill housing under CEQA. “Infill” is pretty broadly defined here: If you’re building a home where there are already other homes around, then you cannot use CEQA to sue on environmental grounds and gum up the works.

These are two huge victories. That said, we could abolish zoning tomorrow and replace the entire permit approval regime with one big rubber stamp and it won’t matter, as long as building new housing doesn’t financially pencil.

So the third reason we don’t have housing, and the one we’re focusing on now, is cost. This can take forms like labor costs being a bit higher here than in Texas or Florida, but what we’re really talking about are government-imposed costs and mandates. Things like unfunded inclusionary zoning mandates, which mean it’s illegal to build multifamily housing in many jurisdictions unless the developer also agrees to provide subsidies for low-income renters or lower-income homeowners.

Or there might be $80,000 per unit in parks fees. A city like Palo Alto, which is literally two-thirds protected open space, will still impose incredibly high fees for parks. These all really add up. Depending on the jurisdiction, you can be looking at upward of $150,000 to $200,000 per home just to get permission to build, on top of already high costs.

Clara: So you’re now more interested in focusing on this last category — these additional cost requirements.

Brian: Yes. Inclusionary zoning is an extra cost. Requiring you to install all the hookups for solar roofs on an apartment building is an extra cost. We’re working on a whole bunch of different cost drivers — whether it’s unfunded inclusionary zoning mandates, transfer taxes, parks fees, various housing impact fees, and on and on.

I’ll give you an example. One of the bills we did recently with Sen. Scott Wiener, SB 937 a couple years ago, shifted the time at which fees are due from the application stage to when the certificate of occupancy is issued.

This might sound like an incredibly small thing, and it is very wonky. But let’s say you’re a builder, and even under good circumstances, the time between when you’re submitting an application and when you’re about ready to lease out the units is several years. Given that these fees, if you’re building a large apartment building, can be well in the millions of dollars, if you have to pay those fees as a lump sum several years before you collect any rent, you’re paying high interest on loans. Instead, if you’re paying those fees right before you actually start collecting rent, that changes the economics significantly.

We’re also looking at building code reform and other types of cost drivers. There are all sorts of things in our building code that may sound like they’re clearly for life or safety, or maybe to achieve some environmental or efficiency goal. But when you dig into it, you often see either various forms of interest group capture or just bad ideas that were codified decades ago.

Subscribe now

Building code chaos

Clara: This brings up my next question. You say “our building code,” but there isn’t one California building code. There are hundreds of municipal building codes. I believe the starkest case is SB 9, the by-right duplex law, which was hit by hundreds of municipalities demanding increasingly specific and arcane building code restrictions on what you can actually do. In an environment as fragmented as California municipal building codes, what are the levers for actually reducing those costs?

Brian: There’s a lot that can be done here. Just last year, the state legislature passed a bill to freeze all building codes. It’s now illegal for local governments to pass any new requirements for building codes because the problem has gotten so out of hand.

Getting to the core of your question: We have a number of proposals in this area, some we’re working on this year, others that are more long-term.

Things like legalizing single-stair1. Things like elevator or “core reform” — the rules around that so we can actually get more affordable elevators into more buildings. Things like requiring the California Residential Code instead of the California Building Code — which I know is confusing, but effectively the codes are different if you’re building a single-family home versus an eightplex. If we can make the codes more like single-family homes for the eightplex, you can actually build them much cheaper.

Quite frankly — and I know this sounds a little bit hacky — rather than try to come up with a bespoke California code that is one statewide code, I would prefer to create effectively an overlay zone where a developer may elect to build to EU code standards instead of the local building code.

Clara: Wait, I want to make sure I understand. There’s one building code for the whole EU?

Brian: My understanding is that there are EU code standards. While I’m not an expert on EU rules, my understanding is that EU law creates functional requirements and governments publish codes for how to meet them. For example, the written code starts with outcomes like safety, health, moisture control, ventilation, energy performance. Then builders can follow the default guidance, or use analysis, testing, or engineering judgment to show other ways of meeting those standards. The regulator’s job shifts from “did you follow the recipe?” to “did you meet the objective?”

Clara: I’m looking out my window in Berkeley right now, which has a different building code than Oakland, which I can also see out my window.

Brian: It’s intensely frustrating when you look at the kinds of projects going up in France or Austria or Denmark or Spain. When I’m looking at the cost per square meter my first reaction is always, “Oh wait, I’m messing up the dollar-to-euro conversion or the square-foot-to-square-meter conversion, because no way is it this cheap.” But actually, the cost to build in those jurisdictions — higher quality stuff, by the way, than we’re building here — is really astounding.

Clara: If you can build according to EU codes — how would that work? Would that be a geographic zone? Would municipalities opt into it?

Brian: No. What we would effectively propose is that of the 500 or so different jurisdictions in California, if they all want to have their own beautiful snowflake building codes, fine, God bless.

However, we should create statewide “alternative compliance pathways” that follow performance-based outcomes that either borrow or adopt the best European practices where they outperform ours. So if a developer proposes to build something in your jurisdiction, they can ignore your code and build to these alternative pathways. That’s the proposal — anywhere in California.

Clara: So you’re not replacing any local building codes.

Brian: Right. And if localities think they can do better than the EU, they’re more than welcome to try. But the idea is that there’s one standardized performance code everywhere.

I’ll say there are other areas in California state law that recognize you need standardization. This is especially true for manufactured homes. There aren’t different rules between Berkeley and Oakland when you’re building with prefabricated structures, because that just wouldn’t work. I would argue that logic also extends to traditional building construction, especially as we’re moving — if not to modular construction — to increased use of building panels and various forms of prefab building elements in factories and then assembling on site. It’s going to be more important to have standardized code adoption.

Clara: And this would ultimately be a state law?

Brian: Yes.

Clara: Do you expect a lot of pushback from municipalities on this?

Brian: Yes, I expect pushback from municipalities. I expect pushback from the fire departments and the fire marshal. I expect pushback from all sorts of folks who are staunch defenders of the status quo. We’re not shy about pursuing what we think is optimal policy in the face of opposition. The tricky thing is, what is the absolute strongest policy that you can get? That typically means having very hard-fought, closely fought legislative battles, because if everyone agrees, there’s probably not much opposition and it’s probably not going to do much.

Clara: So the EU zone is one of your optimal policies. What are the other prongs of this strategy?

Brian: I already talked about some cost initiatives. We’re also working on financing. It’s difficult, especially if we’re talking about building high-rises — steel and concrete in city downtowns throughout California — to get those projects to pencil right now.

What we’re looking at is trying to create a state revolving loan fund where the state is able to provide low-interest loans to projects right before vertical construction permits.

What many other states do, which have had a lot of success, is various forms of property tax abatements. Florida has had a very successful law similar in concept to some laws we’ve helped pass before. I’ll note that the specific program in Florida also required effectively a form of workforce housing for lower and middle-class people. But the cost of that subsidy was provided with a tax abatement, and it’s very difficult to do that here.

The big difference is that developers in Florida are able to get huge breaks on property taxes, and we don’t have that in California. Prop 13 ruins everything. It really makes it difficult for state and local governments to incentivize what we want in terms of building.

The most hated state amendment

Clara: Have you ever seriously considered pushing for Prop 13 repeal, or is that just not plausible? Actually — for anybody who isn’t familiar, what is Prop 13?

Brian: Prop 13 and related propositions effectively state that your property tax assessment may not be more than 1% of your property’s value, and that assessed value can only go up by at most 2% a year. The effect is that you have older Californians who bought their home for $80,000 40 years ago and pay about a nickel a year in property taxes, meanwhile their millennial next-door neighbors are paying $30,000 a year in property taxes. It’s a grossly unfair system.

Until recently, you could inherit these tax assessments without restrictions too. California, which views itself as this progressive bastion, had this sort of system that treated taxes based on your blood relationship to the soil. I would argue blood-and-soil progressivism doesn’t make a whole lot of sense.

Clara: And this has historically been very hard to change. First, because it was a ballot proposition — an amendment to the state constitution — so it would need another ballot proposition to repeal it. And second, because people don’t like their property taxes increasing.

Brian: That’s right. I think there are a few ways to tackle this. I have been working with an academic research institute for several years to see if there are any legs for challenging Prop 13 as a violation of the Federal Fair Housing Act. Long story short, we spent a lot of money and a long time crunching numbers, and it turns out the situation is too muddied — it’s unlikely we would prevail.

We were also part — I’d say a junior part — of a prior reform effort called Schools and Communities First, which would have instituted what’s called a split roll. It would have allowed for easier reassessment of commercial properties, but wouldn’t have changed anything for single-family homeowners. It nearly won.

Now, I think we’re likely back to thinking about what we can do in terms of a different form of split roll, and what is the intersection of “good policy and tractable with reassessing single-family homes, or perhaps different classes of single-family homes, over time. More polling and policy work is needed there before we have any real proposal to take to voters.

Clara: Just to make sure I understand, how does Prop 13 interfere with being able to do the Florida-style kind of tax abatement here? Is it just that we need the revenue?

Brian: No, it’s that it’s so prescriptive on how property taxes are assessed. If you’re providing a discount, how does that property assessment jump up later? How does that end? That’s my understanding. That said, I’ve talked to some attorneys who think there’s more wiggle room here than the prevailing wisdom has it.

The shape of cities

Clara: A lot of the things we’ve been talking about — SB 79, these tax abatement ideas — are focused on larger multifamily buildings. And I get this because I’m also a California urbanist who goes to New York and thinks, “Wow, I wish we had real dense cities like this on the West Coast.” But someone might say, a state like Texas that’s actually building a lot of housing — that’s single-family homes.

Brian: It’s both. Look at the Austin skyline now versus the Austin skyline 15 years ago. It’s totally changed. That’s all dense housing. Houston too. Houston has this reputation for infinite sprawl, which is pretty well-deserved, but Houston has also densified significantly, and they’ve densified in diverse ways. You go from single-family homes to townhomes in many areas — from one single-family home to say three townhomes per parcel — as well as towers, five-over-ones, and everything in between.

If you look at the successful strategies employed in the parts of the country that are growing, mostly in the South, they employ all-of-the-above strategies. They’re building lots of dense multifamily housing. They’re also building lots of single-family homes.

Clara: Would it be accurate to say that CA YIMBY is more focused on multifamily?

Brian: I would say we’re focused mostly on infill, but infill broadly defined. We’re not just focused on building more housing in San Francisco and Oakland. All the places that currently have housing — including suburbs and exurbs — it should be legal to build denser housing there. We’re not just focused on big buildings. We’re also focused on what’s sometimes called missing middle housing2.

Clara: Strategically, what was the decision behind focusing on infill versus, say, greenfield3 development?

Brian: A couple things. One, I would just say values. Most folks in the YIMBY movement want more affordability, more inclusive communities, but we also want more walkable communities and places where you don’t necessarily need a car to get everywhere. The only way to get that is density. There are also serious climate and ecological reasons.

That said, we’re not opposed to sprawl or new development. Some of our legislation varies. Where we’ve pushed for big upzones is in existing heavily urbanized areas. But we’re proposing allowing ADUs everywhere, allowing duplexes everywhere. Our permitting reform bills mostly apply everywhere. It’s not the case that our efforts have only been focused on urban areas. We’re also supporting efforts like California Forever.

Clara: We interviewed Jan Sramek about California Forever in an earlier issue.

Brian: I think that’s a great proposal. He’s trying to build a little bit of Czechia in Solano County, which is great. But I would also say the greenfield developers in California are already organized. They’re well-organized and well-represented in the political system. The California Building Industry Association is their main representative in Sacramento. We work closely with them on policy development and advocacy. The infill folks really aren’t organized. You don’t have an infill lobby in California the same way.

Clara: Why is that?

Brian: There aren’t as many of the greenfield developers. It’s easier to organize 10 people than 1000 people. The infill builders tend to be smaller. They’re hyper-competitive with each other. Historically they haven’t really organized themselves. There’s a group, the Council of Infill Builders, who we work with — they’re great —but it’s just a few people.

Clara: I also want to touch on something you mentioned earlier. You talked about ADUs and duplexes. The ADU legislation you guys have sponsored — huge success. Tens of thousands of units. Over 20% of all permits pulled in California are ADUs. But duplexes, much less so. Why? What’s the difference?

Brian: There was a bill a few years ago, SB 9, that legalized duplexes. We knew when we passed SB 9 it wouldn’t do anything. I told anyone privately at the time: guys, just so you know, this bill has too many compromises. We’ll get a few built, but it’s not going to really do anything.

Also, you can basically do anything you can do with SB 9 between two other acts. With the ADU law, we’re now able to build two ADUs on every single-family home parcel. Do you really even need to build a duplex? Now you can build two ADUs.

And there’s another program we created called the Starter Home Act. It’s a very technical bill, but basically, on the right parcel, it allows you to build up to 10 homes. Between the Starter Home Act and the ADU program, do we even need SB 9? Not really.

So the point of the SB 9 fight was to pass this new program in state law, have that hard political fight, and then over the years pass cleanup legislation to actually make it workable, because we simply did not have the political power to pass what needed to pass at that time. This was analogous to the ADU strategy. The first ADU bill passed in something like 1982 and it did nothing for decades.

Clara: Why do you think there was so much more opposition to SB 9 and the duplexes than to the ADUs? These seem structurally similar to me.

Brian: They are. It is weird. I think part of it was — this was 2021, everyone kind of went crazy in 2021 across a whole variety of domains. The NIMBYs were not immune to brain worms and going crazy.

It may have had something to do with the fact that ADUs were seen as, “Oh OK, it’s like a place to build a backyard cottage for your mother-in-law, or maybe you rent it out every once in a while.” It was sort of seen as your homeowner’s thing, a thing for the homeowner.

Despite the compromises, I think SB 9 was seen as, “Oh, these big developers: — to be clear, the big developers didn’t give a shit about SB 9. Ninety percent of these projects are small. But that was part of it. The realtors really ramped up opposition too for very strange reasons. The intensity of opposition did surprise us. It’s hard to say.

Mr. Smith goes to Sacramento

Clara: This actually touches on another set of questions I want to ask, which is one of my hobby horses. I think that Sacramento-oriented lobbying is bizarrely neglected relative to how important it is. You guys are a very effective California legislature-based lobbying organization. I’m really curious about your strategy for doing that and how you’ve figured out how to operate in that space.

Brian: Before getting involved in politics, I sort of assumed that most state elected officials wanted to just get their name in the paper and go march in a parade. Maybe they wanted some high-paid lobby gig later. They weren’t really serious people, whereas the hardworking advocates were working in the public interest, doing their best against impossible odds.

Then I get to Sacramento and find out that the vast majority of state electeds actually are smarter than you think, and they’re genuinely motivated to make lives better for their constituents. They might have different ideas of what the most important things to optimize for are and how to do that. But fundamentally, I’ve found that legislators — the vast majority — are willing to hear you out, willing to hear your argument, and genuinely want to make things better.

Whereas the advocates are kind of impossible — ideologues incapable of critical reasoning. I think that context-setting is important. I found that a lot of folks who are tangentially involved in politics have way too cynical a view of how politics works and how elected officials think and act. And given that context — that you actually have 120 legislators who have a ton of power in aggregate and are mostly reasonable, well-meaning people — there’s a lot you can get done in that environment.

In terms of “how did I know to properly execute?” — well, I learned a lot on the go. Sometimes people ask me, “What’s this secret hack?” I’m like, “I don’t know. Just have a good plan, hire a good team, and work really fucking hard every day.”

Clara: I would say that none of those are trivial, right?

Brian: No, none of them are trivial. It is hard. Don’t get me wrong — when I started California YIMBY, I had thick curly hair and now look at me.

Clara: This is a text-based interview. For the readers, he does not have thick curly hair.

Brian: Not anymore. It’s hard, but most things in life that are worth doing are hard.

Clara: You had fairly little legislative or advocacy experience when you started CA YIMBY.

Brian: Zero.

Clara: So how did you land on the three-pronged legislative strategy that you ultimately settled on?

Brian: I’ve been thinking about these issues for a very long time, writing about it for myself, but I hadn’t been doing it.

What really happened is I learned a lot by doing. The real origin of California YIMBY — the first bill we did was SB 167 by Nancy Skinner, which significantly strengthened the Housing Accountability Act.

This bill had been on the books since 1982, and it basically says that a local government cannot deny a housing application, or condition approval based on lower density, unless there are specific human health or safety standards. Sounds good, but there were so many loopholes the bill was worthless.

My prior nonprofit sued the city of Lafayette for denying a housing project —technically conditioning approval based on lower density. We lost. And then I thought, well, our approach is just going to fail if we don’t fix this law.

I met with Denise Pinkston, a developer who used to be a planner, who introduced me to her friends who were land-use lawyers. We were at a table and I said, “Here was our case, here’s why we lost, here are my proposals to fix that. What else should we fix in this law?”I took their ideas. I literally copied the existing law into Microsoft Word and rewrote it. Then I gave what I wrote to an actual attorney who rewrote what I wrote.

Then, Denise happened to be friends with Nancy Skinner, who was then a State Senator. She introduces it. I traveled to Sacramento every week on the Capitol Corridor — Oakland to Sacramento — go to committee, advocate for the bill. And I was actually making progress. I thought, “Holy fuck. You can actually just do things. You can actually just pass bills.”

It was this experience that convinced me that serious change in Sacramento is tractable. That’s when I pitched Nat Friedman and Zack Rosen, my co-founders, on: well, if I had an actual budget and an actual team, this is how we pass transformative policy in California. But it can’t just be one guy taking the train to Sacramento every week. You actually need a real comms program, you need to grow and leverage the grassroots movement, you need to build a coalition, you need a political action committee, you need lobbyists, everything else. It’s a whole 360-degree campaign operation.

Clara: In general, did you expect to be able to do more with zoning reform legislation, or did you always know from 2017 that you would need to tackle zoning reform, permitting, and these cost structure issues to achieve a really noticeable impact on housing?

Brian: From the very beginning. If you look at my early drafts from 2017, it was zoning, permit reform, and cost control. That has been with us since day one. I actually had a fourth thing, but I quickly dropped it because I realized how incredibly hard it would be.

Clara: What was it?

Brian: Tax reform, like Prop 13 reform. How do we better incentivize approving homes for local governments? But when you really get into this, you’re effectively talking about having this huge multiparty negotiation between cities, school districts, counties, and the state around sharing various revenue streams. That’s important. Someone needs to do that. But that is very, very hard and was well beyond the capabilities of a small scrappy advocacy organization. I quickly jettisoned that fourth pillar from the initial plan.

Where are the cranes?

Clara: What are your projections for the impact of your work? To go back to the very first question I asked you when we met: when are we going to see the houses?

Brian: We already have gotten a lot of the homes that were made possible through bills we passed. You mentioned ADUs. If you talk to a developer about all these projects they’ve recently entitled, they absolutely used either the Housing Accountability Act or the Housing Crisis Act of 2019, SB 330, another bill we did with Nancy Skinner. They use those tools.

The question then is: well, if all these new homes are made possible by bills we’ve passed, why hasn’t the overall state permitting rate significantly increased?

Two big things. The collapse has actually been biggest with single-family homes, which is an issue we don’t really work on. Single-family home production in California really collapsed after the Great Recession and just has nowhere near recovered.

The other thing is local policy has moved in the wrong direction. This is mostly around cost. If you look at Los Angeles passing ULA, the transfer tax — that’s nerfing housing production. A number of cities have imposed higher levels of inclusionary requirements. They’re increasing various types of housing fees. They’re finding new clever ways, talking to their attorneys, to try to thwart state housing law.

This is a dynamic system. If we had the same local rules in place that we had in 2017, yeah, we would have had a lot more homes built.

Clara: What’s driving the increase in these various NIMBY tactics in cities? Is it responding to state legislation? Is it just increased hostility to building?

Brian: Some of it is absolutely in response to state legislation. Some of it is just that the lefty advocacy organizations have really proliferated over the past decade. They’ve gotten a ton more philanthropic donations. A lot of them also get a ton of money from the government at various levels.

Some are just ideological opponents to market-rate housing. You might say, ‘This is crazy. There’s an avalanche of evidence that building more unsubsidized housing lowers rents, reduces displacement, reduces economic and racial segregation.’ All the things these groups say they care about — the things we’re doing are good for achieving their mission. That’s all true, but there are ideologues everywhere who are immune to evidence. It’s part of the human condition.

They’re demanding things like higher, really unworkable anti-demolition standards if a tenant has lived in a building within the past certain number of years. All these sorts of rules and restrictions in state policy that you don’t have in places like Texas and Florida. It really blunts the impacts of the bills that are passing.

Clara: And is that the main source of delay?

Brian: Another part of this is, well, a bunch of our bills take time to actually have any kind of effect. Some of our biggest bills, like SB 79, don’t fully go into effect until 2031. And then it takes a few years after that. These things take time, which is frustrating to folks who are maybe working on a much faster-paced iterative software cycle. But that’s just not how either legislation or the building industry works.

Also, hopefully within two to three years, we’ll actually have passed at least the frameworks for all the big things we need to massively scale housing production. Then it’s going to be about implementation — really working with practitioners, figuring out where these last remaining little roadblocks are to really scale homebuilding.

I’ll say that last year really was the culmination of a bit of a pivot. From, I don’t know, 2021 through 2023, we passed good bills, but I thought we weren’t really passing the transformative stuff we would actually need to 5x housing production. We’re not going to get there through just delaying the assessment of fees from the application stage to the certificate-of-occupancy stage. It matters — it definitely matters — but we’re not going to fix the problem if we’re fundamentally just getting these small technical bills through.

We did a ton of policy work, made some personnel changes, and thought, all right, let’s just do the thing. Let’s propose it, and if we can’t get it, fine, fuck it. I didn’t start California YIMBY to pass a bunch of bills. I started California YIMBY to rebuild the built environment of the world’s fifth-largest economy along much more affordable, inclusive, and sustainable lines.

The story of CEQA

Clara: I want to hear more about your pivot.

Brian: One of those things was the CEQA infill exemption we got done last year. We wrote this proposal and said, “We all know this is impossible, right? You can’t actually just touch CEQA without labor standards, without a bunch of unfunded inclusionary zoning mandates, without all this nonsense from these lefty nonprofits. It’s just not possible.” As Gov. Jerry Brown put it, CEQA reform is “the Lord’s work.” But the Lord’s work will forever be undone.

I’m like, “You know, fuck it. We’re gonna try.”

I was meeting with Buffy Wicks just over a year ago, and we were talking about this proposal.

Clara: She’s my assembly member. Big fan.

Brian: She’s amazing. She’s great. I’d spoken with her about it before, briefed her a couple months prior on what we were doing. She’s like, “Yeah, no, I know, we should do this, but the politics are so hard.” But neither of us is in this to just pass bills. We’re in this to solve problems. Let’s just propose the bill clean, introduce a clean bill, a clean exemption. And if we have to take some amendments, we’ll fight like hell to keep it as strong as we can, but so be it.

She called me a few days later and said, “Fuck it. I’m authoring the bill. We’re going to do it clean.” And it ended up working. It worked.

Clara: What do you think made it happen? What had changed in the legislative environment that made it work better than you expected?

Brian: A few things. One, Buffy is a very strong author, both by virtue of her position — she’s appropriations chair, a very powerful position in the legislature —and two, she’s a very persuasive and forceful advocate for what she believes in. Having a good author helps.

Clara: I think I remember reading an interview she gave where someone asked, “Are you considering running for Congress?” I think this was when Barbara Lee was retiring. And I think she said, “No, I can do so much more in the California State Assembly.”

Brian: Oh yeah. We talked about this. I was like, “Yeah, that’d be miserable. You’d just be there and get nothing done.” She used to live in D.C. She worked in the Obama administration. D.C. is a nice town, but being in Congress right now is miserable. You’re a backbencher in Congress. That sucks. She can get way more done here. She can pass policy that improves the lives of way more people than she could in Congress.

Clara: As you said, the state level is wildly underinvested in.

Brian: But OK, the other things that changed. I think we’ve done a lot of narrative work. We’ve been changing the way people talk about CEQA. We’ve been focused on this for years, ever since we started as an organization. Other folks have absolutely been making the same argument.

Here’s where I think airport books matter. A lot of people go to the airport. Ezra Klein and Derek Thompson’s Abundance book — I really do think it came at the right moment to convince a number of Dems that we really need to stop passing these unworkable everything-bagel bills that satisfy our interest groups and don’t deliver and just alienate us from our constituents. It’s a bad system.

I do think more Democrats, especially ambitious Democrats that are looking for higher national office, that can’t just satisfy local California constituencies — like a governor — looked at this and said, “Yeah, we’ve got to get something done here.”

Clara: So why do you think the clean bill actually passed?

Brian: As I said, strong author. I think we had a really good plan, we had a great team, we advocated forcefully for it, and we introduced it clean.

Then the governor announced he wanted to put it in his budget, and that kind of came out of nowhere. I remember Buffy and I were talking to each other, and we’re both like, “Did you do that? What? How did Gavin do this?” Because he’s never done that before. He’s signed almost every single one of our bills. We’ve always had a good working relationship with the governor’s office. But he’s never actually intervened on our behalf in the legislature before to really put his thumb on the scale and say, ‘No, you’re going to pass this.’

That really changed everything. That massively strengthened her hand once it went into the budget. But I also want to say — I think we could have passed the bill without the governor’s support, but it wouldn’t have been as strong. Would have had some more of the California everything-bagel flavorings to it. This is where the support of the executive matters a lot.

The next 10 years

Clara: I wonder if you have thoughts on the political culture of California generally. It’s certainly common to see critiques, especially from people who don’t live in California, about it being a one-party state with all this crazy ideological stuff. I don’t know — I’m a California optimist, I’m a California native, I like it here. From having worked so long and so successfully inside the California political system, how do you feel about how the process works and about making it work better?

Brian: I want to be clear: again, the vast majority of state electeds, Republicans and Democrats, progressives and conservatives, moderates — they’re reasonable people who want what’s best for their constituents. The long-running joke in Sacramento is that the most bitter partisan division is between Assembly Dems and Senate Dems — just different caucus priorities that exactly nobody outside of Sacramento cares about or knows about. Most Californians probably don’t even know there is an Assembly.

Look, we work with Republicans. Buffy Wicks had a permit reform streamlining package, and we said, “Look, we think it’s really important that we have a Republican author for one of the bills.” We worked with Josh Hoover. We got the bill through. It was great.

Republicans and Democrats work much better in Sacramento than they work in D.C. It’s not even close. The folks in Sacramento who are the most troublesome are the people who are trying to get to Congress. The folks who are looking to solve problems—it’s not as bad.

Clara: Alright. Let’s say we want it to work even better than it already does.

Brian: A few things. We need better funding for outcome-oriented, evidence-based organizations that advocate for the general public interest. The vast majority of money in state elections is overwhelmingly interest-based. It’s unions and business, basically. Those are the two big groups that give money to state candidates. That’s not true in Congress, where small donors matter a ton.

I would also reform how the committee process works. You have these things called committee consultants. It makes sense in theory — you actually want professional staff on committees that understand issues well and can analyze bills and make recommendations to the chair and others. That all is very good. But I do think there are a handful of committee consultants who have accrued power beyond what is healthy in a representative democracy.

Clara: What’s your most realistic assessment of housing in California over the next couple of years? What do you think’s going to happen? What are we going to see?

Brian: Over two years, I don’t expect a lot to change on the ground. Over 10 years, yeah, I definitely do. I really think you’re going to see the fruits of the legislation—I think you’re going to see the impact of the bills we’ve passed on a 10-year time horizon.

In October 1982, then-Burlington Mayor Bernie Sanders wrote a letter to President Ronald Reagan warning him that nuclear weapons could “destroy all life on the planet” and imploring him to “think courageously and boldly” toward ridding the world of these weapons. Now, in what is likely his last term in public office, Sanders is reprising the existential rhetoric he once brought to nuclear war — this time, for artificial intelligence.

Sanders has recently introduced legislation for a federal moratorium on data centers. Data center moratoria have been introduced in at least 12 state legislatures, so he’s not the first person to do so. But he’s one of the few to ground such a proposal in AI safety terms, and I don’t think that framing is spurious. Sanders’ recent work on AI labor displacement — and his conversations with AI thinkers such as Geoffrey Hinton and Eliezer Yudkowsky — evince a sincere concern for the immediate and long-term risks of AI.

In data centers, Sanders imagines he’s found a vehicle for broader AI governance ambitions. These are visible, literally concrete structures that play on a trifecta of grievances: rising electric bills, anxiety over AI and its corporate creators, and environmental concerns.

However, the bill text is more scattershot than his AI safety messaging suggests. His proposed moratorium would continue until Congress enacts laws requiring pre-release government review of AI products, policies to prevent job displacement, protections against electricity rate hikes, community veto power over data centers, a ban on subsidies for data centers, and union labor requirements. Whatever one thinks of any of these individual items – taken together, it reads less like an AI governance framework and more like a progressive policy grab bag.

I share many of Sanders’s concerns about unchecked AI development, but a pause is not a substitute for actual AI governance. And attempting to address many distinct problems – from safety risks, to labor displacement, to utility costs – through a single, unwieldy solution merely makes it less likely that any of them get addressed. Even taken on its narrowest terms, the moratorium would not achieve its goal.

There’s no plausible path to data center non-proliferation

A data center moratorium is an attempted double-bank shot at AI safety – regulating buildings, in order to constrain compute, in order to pause capabilities. But even if it slowed growth at the margins, it would not be sufficient to constrain AI’s development.

Nuclear nonproliferation has been fairly effective because weapons programs depend on rare materials and other controllable chokepoints. Data centers merely require land, electricity, cooling, fiber, and chips. Some of those inputs can be limited at the margin, especially the most advanced chips. But the basic physical infrastructure of compute is quite fungible.

When Singapore paused new data center development between 2019 and 2022 over energy and land use concerns, the regional buildout of compute continued regardless. Microsoft, Google, and AWS have collectively committed billions across Malaysia, Thailand, and Indonesia. Meanwhile, the Gulf states are treating AI infrastructure as their post-oil economic strategy. Middle East data center capacity is projected to triple from 1 gigawatt to 3.3 gigawatts in five years. Saudi Arabia just earmarked $40 billion for AI investments and declared 2026 the “Year of Artificial Intelligence.”

This is the likely outcome of this proposal as well – facilities will simply continue to be built in friendlier jurisdictions with weaker environmental standards. It may slow data center growth somewhat, but it wouldn’t stop AI development – and we’d lose the leverage that local ownership confers. If the proposal inspired copycat legislation in blue states and localities, the effect would be the same: shifting data centers to more conservative and loosely regulated communities.

The bill tries to preempt that international shift by imposing export controls on computing infrastructure to countries lacking comparable AI regulations – but this raises its own problems. In conditioning chip exports on this set of domestic priorities, many of which have nothing to do with AI safety, we impose a strange form of American-centric, tech policy universalism.

The United States does retain significant leverage over high-end chip supply chains – leverage that, if deployed strategically, could support substantive international AI safety coordination. But that leverage is slowly eroding. In the short run, export controls help us maintain our international lead, and that matters.But as countries like China have demonstrated, in the long run, sustained export controls can accelerate chip efficiency and supply chain independence efforts abroad. We should not spend our narrow lead pushing a hodgepodge of American labor and environmental priorities on unwilling partners.

An overbroad definition of “AI data center”

Data centers facilitate all kinds of digital activity – Netflix streaming, retail shopping, Instagram scrolling, online banking. AI represented about a quarter of all data center workloads in 2025, though that percentage stands to grow significantly.

Despite branding itself as an “AI data center” moratorium, the bill captures far more. Sanders’ definition includes any site used for AI model development at scale, as well as sites that exceed 20 megawatts of power capacity and deliver 20 kilowatts or more per rack. These features are not unique to frontier AI data centers; they also appear in enterprise and cloud-oriented data centers. This bill doesn’t disambiguate between these functionalities – and the average voter won’t either.

It’s supply skepticism all the way down

Read Sanders’s statement carefully, and you’ll notice that he doesn’t lead with the AI safety argument. He opens with environmental concerns and rising electricity costs, then he covers job loss, and finally he mentions AI safety risks.

That ordering reveals the real center of gravity here: general opposition to local construction. Sanders cites Denver’s moratorium as vindication of his proposal, but Denver’s pause was driven by more prosaic land use and electricity concerns, not AGI anxiety. Framing it as a win for AI safety retroactively imports existential framing onto what really amounts to a normal infrastructure fight. That same reactivity has helped produce the housing crisis and obstruct our clean energy buildout – but data centers offer an opportunity to break this pattern.

Data centers aren’t the only new loads coming onto the grid – electric vehicles and electrified manufacturing are also driving demand that requires more generation, more transmission, and long-overdue grid modernization. Many data centers are leaning on gas for near-term power, but data centers could serve as anchor tenants for new clean generation, fiber, battery storage, and transmission. Many companies are moving in that direction.

Industrial projects like these are also prompting pragmatic shifts on decarbonization from environmental groups. The Natural Resources Defense Council (NRDC), for instance, just supported its first nuclear project ever, to power a data center.

A moratorium forecloses exactly the kind of creative thinking these projects are beginning to generate.

Simply govern the actual technology

Even if a federal moratorium were feasible, the question remains – what do you plan to do with the time you’ve bought?

There are many ways elected officials could approach this. Congress could fund an empowered AI Safety Institute with the authority to conduct risk assessments, safety research, and mandatory evaluations of frontier models (going further than the current US CAISI). As AI researchers like Arvind Narayanan and Sayash Kapoor have argued, such institutions must assess risks not only at the model level, but in the context of deployment – making state capacity even more essential.

Congress could also instate aggressive chip export controls (independent of this bill’s other conditions), or pass whistleblower protections to support workers who notice specific dangers to public health or national security. And if lawmakers are serious about protecting workers from displacement – not just providing false assurances that it won’t happen – they should streamline and expand unemployment insurance, and support skilled trades that are labor intensive and more resilient to AI. All of these initiatives would do more to address the risks Sanders is worried about.

In messaging bills like this one, the second-order effects of a bill are the effects, and it is difficult to predict if they will help or hurt the cause of AI safety. A progressive-led moratorium could cement AI safety as a “left” issue precisely when it needs bipartisan legitimacy, or it could prompt Republicans to shift from ambiently co-signing the Trump administration’s accelerationist approach to doing more substantive thinking on AI governance.

It could inspire Democrats to propose their own AI governance and international coordination measures. But whether that’s good or bad depends on whether offices take this bill as a literal blueprint for AI governance, or as a symbolic shot across the bow to the AI labs.

Congress is generally a blinkered institution; lawmakers like Sanders deserve commendation for grappling with more diffuse and long-term risks. But lawmakers must honor that seriousness with remedies that actually address those risks.

Sanders was right in 1982 that the threat of nuclear annihilation demanded a strong response, and he’s right that AI demands one now. ​​But AI demands a distinct approach, one that apprehends the reality of AI risk and the folly of substituting a moratorium for rigorous policymaking. If Congress wants to govern AI, it should. If it wants to protect communities from data center-related harms, it should. But it should not conflate these issues for expediency’s sake.

This article has been updated to reflect new legislation.

Nat Purser is a Senior Policy Advocate at Public Knowledge, where she focuses on artificial intelligence and broadband policy. The views expressed here are her own.