# AI Regulation and the China Question > Source: > Published: 2026-09-19 03:15:41+00:00 1 Many intelligent people say this: The United States should not regulate AI unless China does so as well. Without some kind of agreement with China, domestic regulation makes no sense. There is an important point here, but the argument isn’t right. We can see why once we distinguish among different kinds of “regulation.” 2 Suppose that the question is whether to regulate air pollution from power plants. Suppose that regulation takes the form of tort actions for common law nuisance. We would not say: “No common law nuisance actions until China acts too.” The reason is that there is purely domestic harm, and the common law of nuisance, on optimistic assumptions, optimally reduces it (and on less optimistic assumptions, does more good than harm). Now suppose that regulation would take the form of the Clean Air Act. We would not say: “No Clean Air Act until China acts too.” The reason is that there is purely domestic harm, and the the Clean Air Act, on optimistic assumptions, optimally reduces it, or at least produces (domestic) benefits in excess of (domestic) costs. I am not saying that the Clean Air Act is great or even good (though I like it a lot). The point is that it would not be right to say: No Clear Air Act until China enacts one too. 3 Now turn to climate change. Many people say, and have said, “No domestic regulation of greenhouse gases until China acts!” That argument is probably wrong. Here is why. Purely domestic regulation of greenhouse gases produces “cobenefits” - reductions in air pollutants that cause mortality and morbidity. More fundamentally, *purely domestic regulation of greenhouse gases produces a lot of benefits*. Some of those benefits are entirely domestic: reductions in the domestic harms from climate change. Some of them involve people elsewhere. I think the latter should count. Maybe you don’t, but even if you don’t, reductions in greenhouse gas emissions benefit American citizens, even if other nations do not act at all. You might have a bunch of questions here. Do the domestic benefits of reductions in greenhouse gas emissions justify the domestic costs? Good question. Or: How do we quantify the benefits of reductions in greenhouse gas emissions? Also a good question. The social cost of carbon is meant to answer it. And: Might it be sensible strategy to hold off on domestic regulation as a way of increasing the likelihood of getting an international agreement? Maybe so - but maybe not! Plausibly, the opposite is true: Domestic regulation could be a way of spurring international action. Actually that seems correct, as a matter of fact. (I know a lot would need to be said to justify that claim, but it is probably true.) 4 Now turn to AI. Right now, AI companies could be liable for common law torts - for example, libel, trespass, and fraud. Right now, the AI companies could be liable for criminal violations - criminal solicitation and antitrust violations, for example. It would be difficult to defend the proposition that AI companies should not be subject to the common law, or the criminal law, until China subjects them to similar constraints. The reason is not obscure: US law can properly used to prevent domestic harm. Now suppose that there is an AI Regulatory Commission (AIRC), having roughly the same relationship to longstanding common law and criminal law as the Federal Trade Commission has to longstanding common law and criminal law. That is, the AIRC builds on existing law, but goes a bit, or a fair bit, beyond it. The AIRC might protect property rights; it might protect cybersecurity; it might protect other things that are analogous to what the common law and the criminal law now protect. Such an AIRC might or might not be a good idea. But if t is a bad idea, it is not (mostly) because China has not acted. *The goal of such an AIRC would be to protect against purely domestic harms*. China’s action or inaction is neither here nor there (mostly). 5 The no-regulation-without-China idea builds on a simple set of intuitions: Regulation squelches AI innovation; China is not now squelching AI innovation. If the US squelches AI innovation and China does not, 1) the US will fall behind, to its great detriment and 2) the regulation will not achieve anything, because the same harms will arise. So too in the context of climate change, where the intuition is that if the US reduces greenhouse gas emissions. and China and the rest of the world do not, the US will impose costs on its own people without benefiting its own people. China and the rest of the world will laugh all the way to the bank. In the context of climate change, the intuition is not right — though it is based on something real, and though it points to the immense importance of attending to the domestic costs and the domestic benefits of emissions reductions. The intuition also points, rightly, to the importance of an international agreement. In the context of AI, the intuition is also not right — though it points to the immense importance of *specifying what domestic regulation is designed to do. and of ensuring that it is not counterproductive*. If, for example, domestic regulation, designed to reduce the risk that AI will launch cybersecurity attacks, also stifles AI innovation that would improve AI capabilities in fending off cybersecurity attacks - well, that might be very bad. Here too, the underlying intuition points, rightly, to the importance of an international agreement between the United States and China. But it suggests, wrongly, that unilateral action, by one or another country, is a mistake. 6 The conclusion is that an AIRC is a reasonable idea, and in my view a good one, so long as it can be directed and designed so as (1) to protect against serious domestic risks, (2) not to stifle desirable (domestic) innovation, and (3) not to impose regulation that endangers US interests. We’ve done harder things. It is worthwhile, right now, to work on such a design.