A computer monitor with a parody of a tech company logo floats on the waves while human hands reach up from the depths.

A Rising Tide Lifts All Bots by Rose Willis & Kathryn Conrad / 

Millions of Volunteers Are Keeping the Internet Human. For Now.

Processor, CPU or Chip With the Flag of European Union

How Europe Can Escape a Captured AI Ecosystem

People participate in a march to protest the opening of AI data centers, in Vancouver, British Columbia, Saturday, June 27, 2026. (Darryl Dyck/The Canadian Press via AP)

'Pace the Frontier,' They Say. But Who Sets the Pace?

Citizens and tourists experience and purchase products at an ‘AI intelligent terminal’ consumer center on September 10, 2026 in Nanjing, Jiangsu Province of China. The center features over 200 terminals and related products ranging from smart health and travel to emotional companionship. (Photo by Jiang Wenchao/VCG via AP)

AI Governance Reaches Crisis Point Ahead of Trump-Xi Summit

A demonstrator holds a sign with a message that reads in Spanish; "Evict Peter Thiel" during a protest against the businessman and investor, opposing his growing presence and investments in Buenos Aires, Argentina, Saturday, Aug. 29, 2026. (AP Photo/Rodrigo Abd)

Thinking About the ‘Tech Fascists’ and What to Do About Them

People stand inside the California State Capitol, Monday, Aug. 31, 2026, in Sacramento, Calif. (AP Photo/Jeff Chiu)

Supporting Openness & Safety — Lessons From Recent State Laws

The memorial is seen on the 25th anniversary of the September 11, 2001, terrorist attacks, Friday, Sept. 11, 2026, in New York. (AP Photo/Angelina Katsanis)

Expert Views on Privacy and Civil Liberties 25 Years After September 11

People show the paint on their palms during an event marking Red Hand Day, also known as the International Day against the Use of Child Soldiers, in Bogota, Colombia, where illegal armed groups recruit children, Thursday, Feb. 12, 2026. (AP Photo/Fernando Vergara)

Social Media Helps Recruit Children for Colombian Armed Groups

Water flowing over Srisailam Dam in Andhra Pradesh, India in Aug. 2019. Photo by Sharath G. from Pexels (CC-licensed)

India’s Data Center Boom Should Learn from its Dams

A person takes a picture of the newly unveiled sign for Lake Ontario at Fifty Point Conservation Area in Grimsby, Ontario, on Saturday, Aug. 29, 2026. (Keito Newman/The Canadian Press via AP)

Canadian Hydronyms, Builders, and Other Adventures in Middle Power Tech Policy

The Muse logo is displayed on a smartphone screen placed on a reflective surface onto which the Meta AI logo is projected, in Creteil, France, on September 9, 2026, after Meta launched Muse, an autonomous personal artificial intelligence agent. (Photo by Samuel Boivin/NurPhoto via AP)

The Internet Was Built For Human Agency. AI Agents Are Changing the Rules.

A mouse pointer lingers on button in the shape of a cookie. Photo by: Lino Mirgeler/picture-alliance/dpa/AP Images

Photo by: Lino Mirgeler/picture-alliance/dpa/AP Images

Europe’s Cookie Law Is Really a Law About Surveillance

Image of map and computer language over binary codes on server racks in server room.

Global Digital Policy Roundup: August 2026

Electrical infrastructure with a data center in the background, and a large puddle of water on the ground with a reflection of the infrastructure. 

Electrical infrastructure at a Flexential data center in Hillsboro, Oregon, on May 16, 2026. (Photo by Alex Milan Tracy/Sipa USA)

To Change Data Centers, Flip the Narrative on Sustainability

A major data center project underway in Cape Town, South Africa. Photo: David Harrison/The Continent 

Cape Town is Approving Data Centers with Rules That Still Treat Them Like Warehouses

Students return to classrooms in Tuxtla Gutiérrez as the 2026–2027 school year begins across Chiapas. (Photo by Diego Pérez Gómez / JNA Press/Sipa via AP Images)

Building LLMs? Think Beyond Borders

Parents and teachers held a rally calling on Mayor Mamdani and NYC School Chancellor Kamar Samuels to have a two year moratorium on AI in public schools in New York City, NY. July 29, 2026 (Photo by Steve Sanchez/Sipa via AP Images)

Across the US, School Districts Wrestle With Classroom AI Policy

Deepfake vector illustration. Fake media method in tiny persons concept. Use artificial neural networks to make hoaxes video. Falsified machine learning autoencoders system. Fraud systems development.

China’s Disinformation Operations in Taiwan Are Changing with AI

New York Gov. Kathy Hochul (center), flanked by Attorney General Letitia James (right) and supporters, holds up one of two bills she signed in July 2024 aimed at protecting children and youth from the harmful effects of social media. Photo by Susan Watts/Office of New York Gov. Source

New York Gov. Kathy Hochul (center), flanked by Attorney General Letitia James (right) and supporters, holds up one of two bills she signed in July 2024 aimed at protecting children and youth from the harmful effects of social media. Photo by Susan Watts/Office of New York Gov. 

As Age Assurance Moves Into Practice, What Can Policymakers Learn From New York?

An illustration to depict a conceptual network, with various points connected by strings of different colors. The visualization represents networks, relationships, or interconnected systems, such as in social media, data flow, or neural networks.

Why the UK Needs Better Social Media Data Access to Research Online Harms

is published in partnership with the AI Now Institute. 

, we have argued that Europe's AI ecosystem is subject to the gravitational pull of dominant US players. Even where European AI companies succeed, value tends to flow upstream to US AI labs and, ultimately, hyperscalers — firms that are simultaneously Europe's indispensable infrastructure providers and its competitors.

We also showed how the response from Brussels has been to reach for the tools of the market: boosting supply by co-financing gigafactories or funding consortia to build models, channeling institutional savings into VC and growth equity, paired with deregulation and looser labor restrictions. The hope is that European AI companies, especially in health, manufacturing, defense, or legal, can replicate Silicon Valley's flywheel dynamic, in which successful startups produce exits, exits recycle capital and experienced operators back into the ecosystem, and the ecosystem produces more startups.

The internal logic of all these interventions is the same:

once Europe removes regulatory friction, unlocks domestic capital, and concentrates its efforts where it can plausibly win, Europe will become sovereign

In Europe’s response, AI's trajectory is treated as a natural phenomenon: something Europe must respond to but cannot shape. A tsunami to brace for, or a wave to ride, but never a tide that could be turned.

What these interventions never explain is how they would fundamentally change the incentives in a structurally captured market: one in which start-ups like open-source model platform Hugging Face get bought by Nvidia, Germany's AI DeepL partners with AWS, and neither consumers, businesses, nor the public sector can make meaningful choices at any layer of the stack.

To manage expectations: we do not have a silver bullet — a lucid, innovative policy option that solves these problems in one fell swoop without causing pain in the European tech environment. We wish we did. But in contrast to the policy scripts that deepen Europe’s dependence, we have four that encourage a different way of thinking:

1) Take the market's uncertainty seriously

Two things can be true at once: the AI market is heavily concentrated, and it is also in flux. No technological shift or market correction will, on its own, make Europe a global leader in AI or limit the power of Nvidia and the hyperscalers. But which layer of the stack (applications, models, infrastructure) captures value in the long run is a different question, and one the market has not yet answered. Even the underlying technology is anything but fixed: we simply don't know if the dominant approach to building AI will continue to be based on ever-larger models trained on ever more data and compute, and run on the cloud.

What if open-weight models reach durable parity and models commoditize? What if leading labs respond by moving up the stack into enterprise products and agents? What if inference cost, not training cost, determines where value accumulates? What if the market corrects, and corrects sharply? These are just some of the questions pointing to radically different futures that each need a solid policy response.

We recommend that European policymakers start with the scenario the market demand is 

: customers are turning to open-weight models because they are often good enough and cheaper. Many suppliers, as well, are moving towards this 

This is an early form of what the commodification thesis predicts: that several providers will offer models that perform roughly the same, or well enough for most purposes, and that switching between them will be cheap. There are two versions of how this scenario could play out:

In the first, the model itself stops being a scarce asset, as the cost of using interchangeable models becomes the cost of compute. Instead, value flows to whoever controls the hardware that inference runs on (ASML, TSMC, Nvidia, the memory makers) and to whoever runs models most cheaply at scale as part of services, which today means the hyperscalers (Amazon, Google, and Microsoft). Europe would remain dependent, but the dependence would simply move down the stack, from models to compute.

Arvind Narayanan and Akash Kapur's version

 of this scenario, AI Labs respond to this development by moving up the stack towards applications like enterprise products and "agents" that are embedded so deeply in workflows that removing them becomes prohibitively expensive.

Europe would remain dependent on the dominant labs 

 on the hyperscalers whose compute they come bundled with. This runs directly 

 the interests of enterprise clients and the public sector who want choice and sovereignty.

Both of these scenarios demand a radically different policy response from Europe than the current frontier-access-at-all-cost debate assumes.

Whether models commoditize depends partly on whether large buyers demand portability or accept lock-in (and whether regulation and enforcement enable switching). Whether open-weight models stay relevant depends partly on whether anyone with resources keeps building on them, and on whether public procurement creates demand for them. Whether the labs' migration up the stack ends in entrenchment depends on whether interoperability requirements arrive before businesses and the public sector get locked-in.

In any case, Europe needs to urgently build the capacity to closely monitor where the market is heading, so that policy can prepare and adjust, rather than place a bet on a single possible future. Real scenario planning is not wishful thinking or apologetics — it is a condition for having agency in shaping the trajectory of AI.

2) Actively shape the market to be more open, more competitive, and more interoperable

Whatever AI's trajectory, one thing is clear: the current market is horizontally and vertically concentrated in ways that no single technological shift can undo. A handful of firms, such as Google, Microsoft, Nvidia, Amazon, and Meta and the labs they bankroll, control access to the key inputs necessary for building AI. Crucially, they also control access to consumers.

The money they are pouring into these captive ecosystems functions almost like private industrial policy: subsidized 

 fundamentally steer what the AI market looks like today.

That is why the political debate over whether governments should engage in more active market shaping misses the point. The AI market is already being shaped, not only by governments but by a small number of companies with unprecedented power who are shaping the market in their own interest. The question is not whether the market gets shaped, but by whom.

The very least Europe can do is to stop incumbent firms from entrenching their dominance further: prohibiting self-preferencing for cloud providers and foundation model companies, or interoperability requirements across cloud infrastructure providers, are two examples of where to start. This is classic competition policy. Europe has these tools (although they need updating for the AI market), but they are still not viewed as a key tool for industrial policy strategy.

The more ambitious approach is to use every available lever to shape the AI market in Europe’s favor. Scenarios where AI stays expensive and the market stays concentrated are worse than ones where AI gets more open, more interoperable, and more competitive. But this needs to happen across the stack rather than at the model layer alone. Shaping the market in this direction is in the long-term interest of European companies worried about their margins, European governments worried about sovereignty, and European citizens worried about their jobs and about depending on tech companies they do not trust or cannot control. Shaping the market is not enough to ensure AI truly benefits people and economies (as we explore below), but it is a necessary condition.

None of this is easy. And the European debate treats market shaping narrowly as either subsidies or the occasional antitrust case. But the toolkit available to democratic governments is far wider: procurement, standardization, trade policy, taxation, prohibitions on bundling and tying, coordination among domestic firms, and only then public investment and competition regulation. If Europe is already debating compute expansions of unprecedented scale, it can debate deploying the rest of its toolkit as well.

None of this works if the state is the only actor that changes course.

As long as private actors are overwhelmingly co-opted and vassalized by the existing technological empires, the momentum will not change. What is not often said out loud is that established private companies lack the incentives to jump off the bandwagon. European incumbents resell hyperscaler capacity, take their investment, and sit inside their partnership programs; defecting means forgoing revenue today for a market that only exists if everyone defects together. That's a collective-action problem - precisely the kinds of problems that states are designed to solve.

3) Treat cloud dependence and AI sovereignty as the same problem

Europe's dependence on three hyperscalers and the concentration of the AI market are one challenge, and the relationship between them is the most overlooked piece of the puzzle. 

, most demand for AI is currently for proprietary foundation models that come tied to hyperscaler compute. As a result, European cloud providers face a chicken-and-egg problem: without demand for open-weight models, they cannot justify investing in AI-grade infrastructure, and without the infrastructure, the demand goes to the hyperscalers.

This bundling between closed models and hyperscalers is what makes AI dependence different from the cloud dependence Europe has already known. A database or an email system could in principle run anywhere. What keeps the public sector and so many European companies hooked on the same cloud providers is inertia, long-term contracts, egress fees, and dependence on the services that come bundled with the dominant cloud providers. If you are using a frontier model today, you are still predominantly relying on the hyperscalers, despite a recent wave of multi-cloud diversification. 

While OpenAI and Anthropic have signed massive auxiliary infrastructure deals with neoclouds, their core operational backbones remain firmly anchored to Microsoft Azure, Amazon Web Services (AWS), and Google Cloud. Meanwhile, Google collapses the distinction entirely by providing their models through integrated platforms.

The economic harm of hyperscaler dependence in AI is also different. A company’s (often substantial) cloud bill is still an overhead that stays roughly the same whether the company does well or badly. AI inference costs depend on token use, and once proprietary AI is built into a product or a workflow, the provider takes a cut of every transaction that runs through it. If AI becomes as ubiquitous as its makers need it to be, that means a 

This is why the Cloud and AI Development Act (CADA) fails to tackle the root of the problem. It treats projected future demand for one particular kind of AI, large-scale models run in the cloud rather than on devices or on premises, as a generic need for which Europe must expand capacity. If the paradigm shifts, that capacity sits idle. And if it does not, the demand in a captured market is predominantly for closed models that come bundled with hyperscalers.

CADA only reserves a fraction of public sector demand for European cloud providers, while aiming to triple data center capacity in ways that will predominantly benefit the hyperscalers. All of this leaves the fundamental structure of this market untouched. Especially in scenarios where AI commoditizes, this is fatal, as it concedes one of the layers of the stack where money is being made. Just not by Europeans, but by an oligopoly of cloud providers that can continue to seek rents.

To avoid this future, Europe needs to treat the diversification of its cloud market, not its generic growth in size, as a key goal of AI industrial policy. The cloud-AI dependency also shows why Europe has a strategic interest in AI trajectories where most uses of AI don’t need a hyperscaler at all: open models, or smaller models that run locally, on device, or on-premises. It also shows why procurement is such an important lever in creating demand that European suppliers could meet.

Here is an important caveat: even a market in which AI is cheap, open, and the layers of the AI stack are contestable, is not necessarily one in which AI benefits the public, or the broader public interest. Market shaping contests who captures the value, not how it is distributed, or what is truly valuable in the first place.

Scenarios like Europe 2031 simply assume that deploying ever-larger frontier models will produce economic benefits that exceed the cost of the deepened dependence they bring. That is not just a technological prediction, but an economic bet. Only if you believe that we are 

 (or at least that models will keep improving and remain the cost-effective choice) does this bet seem safe. In that world, Europe’s AI industrial policy becomes about negotiating the least-bad terms of its dependence. If you do not, then the question of whether the bet is worth taking is open, and what we have argued throughout this series is that AI’s economic benefit for Europe, at least under the conditions of the current market, is anything but guaranteed. In fact, AI could end up both concentrating and extracting wealth from Europe to the privileged elites of Silicon Valley.

Declining the bet thus requires more than just aggressively shaping this market. It also needs an affirmative vision for AI’s role in society. Europe arguably has more room to imagine alternatives than the US, where the economic momentum hangs on the frontier cycle being true. 

 estimates that making the spending on AI infrastructure profitable will require $2 trillion in annual revenue by 2030. That revenue has to come from somewhere, which is why we are seeing a push to integrate AI everywhere — even in places where AI tools are not fit for purpose — to see what may stick as a revenue opportunity.

A starting point for a better paradigm could be the distinction between augmenting and replacing. In a replacement logic, LLMs replace workers. This is implicit in the hopes of trillion-dollar valuations of the private AI labs — to replace, capture, and extract value from a whole segment of 

. Augmentation instead means adopting AI services only when and if they truly improve the productivity of workers and help them deliver useful services. Instead of replacing a customer service agent, LLMs can expand what that agent can do by providing new capacities. Choosing between these two ends is a question of 

This is also what decentering AI means in practice: refusing the premise that AI is an exceptional technology around which all other politics — labor, sustainability, democracy — must bend. Instead, it is a technology that can, and must, be shaped around the constraints of what kind of societies we want to have.

None of what we propose requires Europe to own every layer of the stack. It requires enough control, interoperability and bargaining power that people, companies and the public sector can switch providers and make meaningful choices. It also requires accepting AI’s fundamentally uncertain — and thus shapeable — future. AI is neither a wave nor a tsunami. To stop drifting, Europe needs to decide where it wants to go.

Audio of this conversation is available via your favorite podcast service.

On Friday, September 11, I moderated a discussion at the 

Institute for America, China, and the Future of Global Affairs

 at the Johns Hopkins School of Advanced International Studies in Washington, DC as part of 

 called “Getting Tech Right: US China Policy for a New Era.” Amid a frenzy of headlines about AI safety and just two weeks out from a planned meeting of US President Donald Trump and China’s President Xi Jinping, the discussion was organized around the question, what does AI governance actually demand right now? Joining me were:

, SAIS ACF and the Center for a New American Security

What follows is a lightly edited transcript of the discussion.

My name's Justin Hendrix. I'm CEO and editor of a site called Tech Policy Press, which focuses on issues at the intersection of tech and democracy. I'm normally based in New York City. And when I talk about things happening in the specter of the US Capitol, it's not literally right above my shoulder. So this is a great pleasure to be here and I'm thankful to the organizers for inviting me to help moderate what I think will be a very great session today. 

And I'm also planning to use the audio from this conversation for the Tech Policy Press podcast. So I have a kind of habit on the podcast of making sure we do an audio check and let people announce themselves, say their names out loud. That helps me to avoid butchering pronunciations when I do my podcast recording. So I'm going to stick with that and I'm going to just go down the line and ask folks to state their name, their title, and their affiliation so that everyone knows where you're hailing from.

I'm Miro Plueckebaum. I'm the strategy and research manager at the Oxford Martin AI Governance Initiative. And I also run a Singapore and London based technical AI governance research lab called SASH. And

I'm Kat as in meow, Duffy as in Duff Beer on the Simpsons. And I am a senior fellow for digital and cyberspace policy at the Council on Foreign Relations. And I'm also the director of our LEAD AI initiative.

I'm Amba Kak. I'm the co-executive director of the AI Now Institute.

Jack Shanahan, retired United States Air Force, but thanks to Afra, I'm not going to say that anymore. I'm an OPC. I'm a one person company. So I'm very proud of that. I am a senior fellow with ACF, but also an adjunct senior fellow with the Center for a New American Security.

Well, I'm looking forward to this conversation today. We've got a range of perspectives, but I felt maybe I'd start with just reading a couple of headlines and I almost just randomly chose these from my newsfeed. But one of the things that folks have been talking about in the communities I travel in has been just this sort of frenetic pace of news, the frenzy of information, the kind of wild set of events that have taken place over the last couple of weeks, which if you cover tech and you do journalism, at any point in past, any one of these stories would've seemed as if it would occupy days or weeks or months of discourse, and yet they sort of fall away almost as instantly as they're published. But headlines like the Wall Street Journal, “Anthropic says Iran used its American AI model to target US Navy warships.” The Financial Times, “Houthis used Anthropic AI to try to build ballistic missiles”. CNBC, “Extinction warnings ramp up as more OpenAI, Anthropic researchers join calls for an AI slowdown”. And then maybe a little closer to the Capitol here. Politico, “Hawley presses OpenAI over rogue Hugging Face hack.” And finally, Wall Street Journal, “Congress is suddenly waking up to the AI doomsday threat.”

I literally could have drawn any number of headlines like this from the last couple of weeks. But the organizers, if you go back to what they wanted from us for this panel, they wanted us to offer a kind of a possible affirmative vision for AI governance, what we should do, what AI governance demands now. And yet we know that just down the street, senators, Congress, congressional representatives are apparently in a flutter. They're putting out bills rapidly. They're trying to draft legislation in some cases in response to these headlines.

So my first question for you all, and in fact, I'd like to put China out of our minds for just a moment. If you were to walk down there, walk down the street, short clip and tell folks what they should be doing to get AI governance right, right now, what would you emphasize? What would be the thing that you would say should be the kind of framing thought? 

And perhaps Kat, I'll put you on the spot to go first.

Yeah. For a year, I've had a series of AI agents I've consistently been refining to help pull in all the news that I need to see. And now even the agents can't keep up with the news cycle. They're like, "You only wanted me to surface 15 things. What do you want me to do?" I mean, I think putting China out of the equation, the most important thing I think we should be asking ourselves is what is America best at? If America is going to lead in this technological transformation, we're arguably going to lead best by relying on our strengths and leaning into those strengths. And historically our strengths have sat in openness, in due process, in accountability, in supporting innovation, in supporting incredible influxes of talent and in recruiting that talent in. And so I would say that a lot of what we needed to be doing with AI governance right now is sort of returning to the fundamentals.

I think American citizens need to have a better understanding of the way that this technology is being developed. I think they should have a much better understanding of the way that it's being financed. We have massive amounts of opacity in our private capital and venture capital flows in private equity. Americans don't know who's funding what, where that money is coming from, nor frankly, do many of the innovators know who is funding them or who is funding their work. There's no transparency on this at all. Americans don't know what their tax dollars are going to and what they may not be going to. I think when you see the concerns right now about data centers and the impacts of those data centers, you're also seeing, for example, the data center in Memphis that was really producing terrible results for the populations close to it, that was also a result of permitting processes that had been shortcut.

So it wasn't that the permitting process didn't exist or that the governance didn't exist that would've made that data center more safe for the community. It's that there was such a rush that the permitting and the governance that was already in place to protect citizens was thrown by the wayside on the altar of supporting this technology. And so that's where I would say let's return to where we're best to begin with.

Okay. So transparency and embracing some of those principles that have made America America in your view. Amba, what about you? What would you emphasize to the folks down on Capitol Hill right now what AI governance demands?

Can I do one second of throat clearing first, which is I think there's at least one person in the audience that agrees with me. I can't see him anymore, but it feels like every time Sam and Jessica, you organize these conversations, they really do feel like a refuge from every other conversation happening outside this room. The most optimistic note I guess I can strike at this moment is that it does feel like the distance between the conversation or the kinds of discourse we're trying to advance in this room feels less further away than it has at many points before. I was thinking this morning that when Ted Cruz comes out in the midst of this news cycle to say that they'd rather it be our killer robots than China's, I do think we've got - 

He needs to think about that one just for a second.

It didn't land well. It didn't land well. I think there is a visceral mainstream backlash to just the ridiculousness of the race to the bottom logics that have been perpetuated around this frame. And so that's the good news. Maybe the flip side is you never waste a crisis is maybe the most important policy adage, but equally I think not every policy crisis is equal in the opportunity it creates. And what I'm sitting with in this moment is that the nature of the extinction risk whistleblower moment that we're in right now, it seems like all roads are in fact leading to a policy imagination that looks straight back at the industry for answers to problems that they are creating. And that is what maybe gets us closer to, are we going to waste this crisis or even worse, are we going to undercut all of the public momentum that has come not just from tech whistleblowers, but from data center activism, from parents groups?

Is that going to be undercut towards a policy regime that is essentially functioning as a checkbox certification legitimization of business as usual rather than real accountability and real oversight? And there's many ways in which this is happening. I think the fact that the Bernie Sanders Kassar bill hinges on a definition of super intelligence that is frankly undefinable and would leave most of industry off the hook is a good sign, like the smell test that something is off with that line of that sort of solution being what we see from this or what we make of this crisis. Okay.

So a lot of threads there, but put super intelligence out of our minds, focus on what the public needs, the public interest rather than the corporate interest. Lots there, but we can come back. Miro, what about you? What does AI governance demand now maybe with your verification hat on?

Well, the examples you gave to kind of set us up already illustrated that the stakes are really rising. AI systems are becoming more powerful. The harm they can do is increasing, but so are competitive pressures between companies. The prize is starting to feel quite real now, regardless of what you think of the valuation of companies being overvalued or not. The valuations are no longer imagined. They're actually generating quite a bit of revenue. And what that means is that in cases where companies do want to behave responsibly, they may get penalized by the market. So you mentioned briefly the Hugging Face incident in the introduction. In that case, OpenAI tried to act responsibly and pause the training of the model for a period of two weeks. Now in this case, they chose to do it unilaterally. No other company paused, but we can't be living in a system where the responsible actor can't behave responsibly because they have to worry about their competitors undercutting them.

So if I were to walk down the street, I would ask folks to start thinking about what type of a regime can we build that allows responsible actors to behave as such when required without being undercut by actors that may not want to behave as responsibly. And that's an open policy question. We're focusing on building tools that can verify whether or not actors have, for example, actually paused, but at what point we ought to pause and for how long and what others should do is a question that we need folks down the road to help figure out.

Okay. Jack Shanahan, I know you've worked directly on these questions around US and China, but again, keeping China off the table for the moment. If you were to go down the hill, AI governance, what does it demand now?

Well, first thing I'll do is go back to ground zero, so to speak, on September 11th and redefine what do we mean by governance in the first place? To me, it's rules of the road. I spent 36 years in the military. I can't remember ever using the word governance, but governance ruled everything we did. We called it policy, we called it directives. What are you allowed to do, what you can't do, what you can do under certain circumstances, who is held responsible and accountable when something happened? All those things were what I call rules of the road. We need that in AI in a way that we just haven't had before. This will be a little bit provocative. I'm not sure people would disagree with it, you might take exception with certain ways. I'll say this, but at the end of the Cold War, we entered into a period of unfettered technological libertarianism.

The government was not in the equation with what was going on in Silicon Valley. The government wasn't the customer for these companies. They raced out in front of everybody else. The government didn't understand the technology, had no interest in figuring out how to govern it properly. And the next thing you know, here we are today and people saying, we blinked twice, we look around and this technology's outraced all of us. So now is the time to reset and say, what are the rules of the road? Some of you may be familiar with the term Collingridge dilemma. I am very sympathetic to the Hill in many respects is you move too fast on a technology and try to put out policy, you move too fast, it's the wrong policy and you screwed something up. You wait too long, the technology has taken root. Now what are you going to do about it?

It's really hard to get it in past a certain point. I think people are very serious on the hill about trying to find where is the right time to come in. I think we all sense that that time is now. These incidents are just making it clear to everybody, if you don't act now, something serious is going to happen and we're all going to regret that. So for me, it is a very basic term. It is rules of the road. That's a very easy statement for a lot of different things that have to be done to get it right.

So I want to pick up on the point of time. It's really the second thing I wanted to get to a little bit. The question has been somewhat, is there going to be some major event that finally gets everybody off the couch and to take action? It does seem like right now maybe there are lots of important things that are happening or frightening things that are happening that together people are beginning out there in the real world. Normally people who don't think about AI policy every day are piecing together in their mind and they're building their own imagination for is this a crisis? I don't know. How long do we think we've got to take advantage of a sort of push towards there needing to do something and how do we avoid doing something when perhaps in some ways we should have done nothing? How do we avoid making mistakes in this moment?

Does anyone have a general view on that? How do we take advantage of the moment without somehow, I don't know, making a mistake in the process?

My initial reaction when the first part of your question is when will that crisis arrive or when will we be there? And I think we're there. The question is no longer are we there yet to the crisis that opens up the political window. The policy window is before us. And maybe the best evidence of that is the fact that with, I think no exception today, every single AI firm is standing up to support in theory the prospect of regulation. The terms though are the ones that I think the fight right now is about who is going to sort of set the terms of the debate. Is it, as Jack is saying, going to be this sort of amorphous notion, to me, a softer notion of governance which really is often self-governance in other clothing or is it going to be much more straightforward leverage that the public can actually assert over this industry?

And I think that looks like rules. It looks like liability. It looks like the hard, the concrete enforceable rules. It also looks, to be honest, for one, let's start with making sure we're not wiping off the rules that do exist off the books because that's as a sort of reality check. We're at a moment where in the last few years we have seen AI exceptionalism that I think no industry has had it this good in some ways. So it isn't just no new rules for this industry. There have been several attempts, moratorium in many different shapes and forms, pushes to try to immunize this industry from even adhering to the law as it exists on the books. So the DOJ is stepping up in the New York Times case, also in the Memphis permitting case, and their argument is China, that if the US is to hold national security advantages, then whether it's unpermitted turbines in Memphis or a copyright suit, you cannot touch the AI industry.

And I feel like that the moment to reckon with or challenge that premise is here, but my sort of anxiety and also optimism is like, can we make the best go at it?

If I could jump in, I want to go back to something you said at the very beginning about this time element where you read these headlines just from the last, what, 48 hours, 72 hours. What makes this technology so different than I think with anything any of us have experienced in the past is the time you would have to actually work on things where now you've got this accelerated rate of development, this breadth and depth and diffusion and this rate of adoption that is so different than anything before. It feels like that yes, the time is now to do something, but the pace is so brutally relentless that human nature says, no, no, give us some time to focus on the headlines you just read. Then the next thing is about to happen. And maybe it's only when that catastrophic event happens that we really do see action.

We wish I could be, I know you want to leave on an optimist note, so maybe we'll get there eventually, but at heart here, the signs are not really great for solving these problems right now with regulation until something else really changed, whether that's the election, whether it's a catastrophic event. But I think this time pressure is different than technologies that we have experienced in the past.

Kat, Miro, this time pressure, anything to say?

I mean, look, if you're on the hill, your time pressure is the midterms, right? I mean, if you're looking to people to govern over the next few months as opposed to run, I wouldn't hold your breath.

And so it may be less the crises that should be responded to because those aren't necessarily the crises that the electorate is going to hook to as being their kitchen table issue, their thing that they would go to the ballot for. What's really interesting is the polling that's been coming out that is showing that unprecedented amounts of agreement across evangelical Christians and liberal progressives absolutely polling at the same numbers and like 80% use of AI in K-12 education. There are going to be these issues that I think when those of us who are foreign policy thinkers or operate in the NATSEC space and cyber, those other areas, I think are underestimating potentially what Americans are going to be most concerned about. And that type of polling in the lead up to elections between that and the data centers, if I were saying where I think governance might sit or governance proposals might sit, I suspect it will sit in wherever electability sits, not necessarily in where the triaging actual crisis does sit or should sit.

I'm not as policy savvy as my fellow panelists, but just one observation that I'm going to make in a personal capacity in response to the way you framed the question, you said, how do we avoid doing something wrong? And that implies some sort of hesitancy over what to do from the policy point of view. And it's not something I sense a lot from the private sector. The adage is move fast and break things. And on the policy side, it seems to be much more hand-wringy and hesitant. And if policy wants to keep pace with technology, it needs to perhaps move into a much more fast moving, iterative direction where it is understood that you're just not going to get it right every single time, but you need to be able to adjust very quickly on the fly in response to a changing landscape as the companies are doing.

I mean, they're very much playing to win. They're not just. Yeah. So that's just a quick observation there as someone who's, again, not very policy savvy. Well,

And I think it's a good observation and those of us who've worked on tech policy issues, perhaps one of the things that's working against us is that we are convinced that if a law passes, for instance, we should be prepared to live with it for 40 years because nothing is going to change in the intervening time. That seems to be what we all operate on. But let me now kind of bring the China question back in because Amba mentioned, for instance, the DOJ's effective support for Elon Musk's xAI in Memphis, the argument effectively that we can't really expect to hold xAI to account for poisoning the air in this Black community in Boxtown in South Memphis because we've got to beat Beijing. Because

That's right. Let's bring China back in. Let's bring the question back in. How does that change anything about your prior suggestion about what AI governance demands now? How does that inform? And Jack, with this one, I'll start with you.

This is a conversation Paul Triolo and I were having last night based on watching his podcast last week with Kendra Schaefer on this very topic is to have this governance conversation between the United States and China, which I think is absolutely essential. I can't question that question is what does it look like? But to do that, we have to get our own governance house in order and China has to get its own governance house in order. This is not arms control is owned by this one organization in respective countries. It's AI that brings in everything. It brings commerce, it brings treasury, brings in defense, it brings in state. Keep going down the list and the equivalents in China. So first of all, it's a very complicated, messy piece to go down the road of who are we talking to? Why are we talking about it? Exactly what are we going to be talking about?

These are issues that I hope are being resolved. I'm not seeing a lot in the public discussion about on the road to the 24th September meeting, but there is no better time to get going on a conversation. Why would two AI superpowers, do these two AI superpowers want to have this conversation in the first place? I think you start there. We have mutual vulnerabilities and mutual risks when it comes to AI. Without those, I think all you're doing is having trade-based conversations about trade imbalances, tariffs, and all the things that we have been doing vis-a-vis US and China. Now with this technology is showing capacity to do things that are troubling people globally, there is no better opportunity to get going on the conversation, but you can't do that unless you get your own governance house in order. I don't know if we have that in order yet.

So if I were to say worst case, it's actually not the worst case. The worst case is not talking at all. The sort of less worse is we just go in and get started. Okay, that's all right as a starting point, but it can't be a one-off conversation. This is going to be an institutional dial. Let the track twos do all the work that track twos and many people in this room may be familiar with or involved in track twos. They're there to do a lot of the homework before the track ones get going. In this case, I guess the last thing I'll say is track one by itself isn't going to be sufficient because these are technology companies, unlike any other technology that we've worked in the military in the past. The technology companies have to be involved in the conversation exactly how they are, who they are.

Those are big questions that have to be addressed both in China and in the United States. But I don't see how you have a very fruitful conversation on AI governance without the technical expertise to go along with the policy expertise. Those have been two different circles that need to be a Venn diagram before we get into really good conversations on governance between the US and China.

So maybe before we move on from you, how confident are you that there's a lot of folks who are wondering about what will happen in the next couple of weeks. It's not even clear. Is it clear yet if we're actually having an AI dialogue? We don't really know, do we?

I don't. I hear rumors. I hear things as they say in New York. I hear things. Something will come out of this. So there is no question. I don't think there's any question. AI will be on the agenda when the two presidents meet. How much will be done in advance of that? I feel not so great about that, although some people are telling me, nah, some work is going on, you just don't see it. Okay, that's fine. Now we'll see what happens. More likely to me is September 24th comes and goes, then there's an agreement for the two sides to get together. Now you start looking at what are the parameters of the dialogue? Is there a separate mil-to-mil talk? I mean, all those things can be worked out. I just see the appetite still is there, but the release of the distillation report followed by China's reaction to that felt like, oh no, are we about to say talks are off?

I don't think that's the case. People are telling me, no, no, they're still on track, but I will be surprised. If I'm wrong, please blast me on LinkedIn or somewhere. I'd love to be wrong. I don't see something really substantive happening before September 24th. Today is the 11th. We have two weeks.

Kat, you move in similar circles in New York international diplomacy. What are you hearing?

I concur entirely. I think another way that I've been thinking about it is what could come out of the dialogue on September 24th where that announcement or the execution of whatever was promised, the entire world would essentially be in agreement. Xi and Trump deserve to share a Nobel Peace Prize for that. What would win them a joint Nobel Peace Prize? Where the entire world would be like, thank God that they work together on that. We are all safer. We are all better for it. And to Jack's point, there are mutual interests that sit exactly in that category. What is coming out in terms of biohazards is terrifying. All of the work that happens in CBRN, in chemical warfare, what's happening with our lack of ability right now to do truly comprehensive good faith threat sharing among frontier companies. And it can't just be the American frontier companies.

We need true cooperation in key areas. And that is, I truly wish that you could take Trump's top three advisors, Xi's top three advisors, put them in a room and say, think of everything that could win them a Nobel Prize. Start the conversations there and then move backwards.

Amba, can you see that Nobel Prize coming?

No, I can't claim to be an expert on that piece of it.

The Tech Policy Press newsletter is published every Sunday. 

Subscribe here to receive it in your email inbox

People participate in a march to protest the opening of AI data centers, in Vancouver, British Columbia, Saturday, June 27, 2026. (Darryl Dyck/The Canadian Press via AP)


The frenzy of headlines about AI safety over the past several days has made it difficult to keep up. From countless items in the Anthropic 

 that spawned headlines that in years past might have eaten up an entire news cycle on their own (see “

Anthropic Says Iran Used Its American AI Model to Target US Navy Warships

Houthis used Anthropic AI to try to build ballistic missiles

Anthropic blocks possible attempt to use AI to make biological weapons

 by a swarm of OpenAI agents (this one took place before the now famous Hugging Face incident), it appears to many to be a 

, as the destabilizing effects of releasing these products into the world without constraints become apparent.

But perhaps nowhere is the discussion over AI ‘doomsday’ more pronounced than 

 themselves, where executives and developers are expressing their vivid imaginations about what might happen next. On Saturday, Anthropic founder and CEO Dario Amodei published another in his series of occasional blog posts to address the crisis, “

.” “We must slow the pace at which we improve the capabilities of AI models,” writes Amodei. “To be clear, pacing does not mean halting model training or technical progress, but ensuring companies take adequate time to align and safeguard their models, and for third party evaluators to confirm this,” he writes.

What does he propose specifically? First is “giving ongoing, employee-like access to a team of embedded third-party evaluators,” which he says Anthropic is “unilaterally committing to” now. Second is coordination in democratic countries “to establish common safety standards as well as limits on the rate of unchecked AI progress.” And third is that the US and other democratic countries should “attempt to coordinate with authoritarian governments, to the extent this is possible, while taking seriously the challenges of verifying compliance.”

In response to Amodei’s post, tech leaders including OpenAI founder and CEO Sam Altman and his nemesis, SpaceX founder and CEO Elon Musk, 

. Another person to generally endorse the call was David Sacks, the investor, former Trump administration AI and crypto czar, and current co-chair of the President's Council of Advisors on Science and Technology. “If the unreleased models are scary enough that you think you should slow down, I support your decision to be responsible,” wrote Sacks in a post on X. But, he added:

But stop pretending you need anyone else’s permission. Stop pretending antitrust law has to be suspended so you can form a cartel. Stop pretending you need a regulatory approval process that supersedes product liability. Stop pretending METR is independent when it is intertwined with Anthropic’s investors and staff. Stop pretending you need those same evaluators to police competitors who aren’t even at the frontier. 

Most of all, stop pretending the motivation to slow down is purely altruistic. You face massive product-liability exposure if your products enable a truly damaging cyberattack. The market already punishes models that behave in unpredictable or unauthorized ways. After the Hugging Face episode, it is simply good business for OpenAI and Anthropic to trade some raw power for reliability and predictability. Call it alignment if you want. It is also just giving customers what they want.

Sacks’ perspective may not fully align with that of President Donald Trump, who says he is 

 and is focused on staying ahead of China. And it is unclear what, if anything, may happen in Congress, where lawmakers are 

 to advance legislation before the US midterm elections. (House Speaker Mike Johnson, R-La., 

 it is the duty of companies, not Congress, to make AI safe.) Either way, tech politics has gone mainstream, as former 

 to make AI a central concern if the party wins the House in the midterms, and in the 2028 race for President.

In the near term, though, another open question is the extent to which these issues will be discussed when (

) Trump meets with China’s President Xi Jinping in Washington 

. That planned summit was in the fore of discussion Friday at an event I attended in Washington, D.C. at the Institute for America, China, and the Future of Global Affairs at Johns Hopkins SAIS on Pennsylvania Avenue. I moderated a discussion (

Oxford Martin AI Governance Initiative, and 

from SAIS ACF and the Center for a New American Security.

We covered a lot of ground in the discussion, which was focused on what AI governance demands in the immediate term. 

 made the case for returning to American strengths in transparency, due process and accountability. 

 warned against letting the current crisis produce checkbox certification rather than real oversight. 

 discussed his work on technical tools to verify whether labs and states are doing what they claim. And 

 argued for the necessity of defining "rules of the road" before turning to what a Trump-Xi meeting on AI could realistically produce. We also discussed the 

 and what it shows about the potential to collaborate on cross-border threats, China's own comparatively aggressive AI regulation, the fraying of the arms-race narrative, the role investors and stock exchanges might play in governing the industry, and more.

We’ll have more on these issues in the week ahead. Read on for more from Tech Policy Press this week.

Based on interviews across 24 countries in 7 regions of the world for a report from the Berkeley Human Rights Center, 

 call on AI developers conducting evaluations to consider how geopolitics, resource and access disparities, and other factors 

Internet governance and AI governance communities have largely stayed separate. 

, senior resident fellow at the Atlantic Council’s Democracy and Tech Initiative, and 

Humane Intelligence Public Benefit Corporation, make the case that they must come together to 

 for an internet increasingly driven by AI agent interactions. “The Internet did not become a global public resource because it was designed for autonomous machines. It became one because its architecture allowed different networks, technologies and communities to interoperate while leaving room for human choice,” they write.

 don't have lobbyists. Tech policy strategist 

 on why that capacity gap will decide whether the next wave of state AI bills protects openness or accidentally kills it. “The lesson is not that “open” should be exempt from regulation, but that obligations should reflect an actor’s actual knowledge and control rather than broadly requiring developers to monitor and control downstream uses,” he writes.

The scale, speed and coordination of AI-assisted disinformation are outpacing the mechanisms designed to counter them. 

 a Fulbright Taiwan Fellow pursuing a master's degree at Columbia University's School of International and Public Affairs, writes that 

, an associate professor in Carleton University’s School of Journalism and Communication and an academic director at Re.Climate, and 

, a professor in the Department of Communication at the University of Ottawa, examine how Canada's current 

Latest

Podcast

Recent

Our content delivered to your inbox

Thank you!

Around the globe