OpenAI just proposed something that would have been politically unthinkable two years ago: a global AI governance body that includes both the United States and China.

The proposal came from Chris Lehane, OpenAI's vice president of global affairs, during a briefing in Washington. Lehane suggested the US should use its current lead in advanced AI to establish an international framework for safety standards — and pointed to the International Atomic Energy Agency as a model.

The timing is not accidental. Donald Trump is in Beijing right now for his first state visit to China in nine years, and AI is expected to feature prominently alongside trade, rare earth supply chains, and Taiwan. The message from OpenAI is clear: treat AI like nuclear technology, because it might be just as consequential.

What They're Actually Proposing

Lehane's vision involves linking the US Commerce Department's Center for AI Standards and Innovation with the growing number of AI safety institutes being established worldwide. The framework would include shared global standards, regular oversight, and coordinated safety testing between nations.

China's participation, he argued, is not optional — it's necessary. The logic is simple: if the two largest AI powers don't coordinate on safety, everyone else is left guessing about the capabilities and constraints of the systems being built on both sides.

OpenAI has also been pushing for government-led testing of powerful AI models in classified environments before public deployment. The company believes independent evaluation could identify dangerous behaviours or vulnerabilities before systems reach users.

That approach may clash with the Trump administration's reported preference for voluntary cooperation over mandatory oversight. But Lehane appears to be betting that the geopolitical stakes are high enough to override the usual regulatory hesitations.

Why This Is Harder Than It Sounds

The IAEA analogy is seductive but imperfect. Nuclear technology is relatively well-defined, physically constrained, and difficult to build in secret. AI is none of those things. Models can be trained in data centres that look like any other server farm, weights can be exfiltrated on hard drives, and the same system that writes helpful code can also write malware with a minor prompt adjustment.

There's also the trust deficit. US AI companies — including OpenAI and Anthropic — have increasingly complained that Chinese developers are distilling outputs from American frontier models to train rival systems more cheaply and with fewer safeguards. If you're already accusing the other side of free-riding on your safety work, asking them to join your governance body requires some diplomatic gymnastics.

And then there's the question of who speaks for whom. OpenAI is a private company proposing a framework that would bind nation-states. That's not inherently illegitimate — corporations have shaped international standards before — but it does raise questions about democratic accountability. A governance body designed in Washington by OpenAI and approved by the Commerce Department is not the same thing as a treaty negotiated by elected representatives.

The China Question

Including China is the most politically sensitive part of the proposal, and also the most strategically significant.

The US and China are in a full-spectrum technology competition. AI is arguably the most important front. Both countries are racing to develop more capable systems, attract talent, and secure supply chains for the specialised chips that make it all possible. The idea that they would cooperate on safety while competing on capabilities is not unprecedented — the US and Soviet Union managed it on nuclear weapons — but it requires a level of institutional trust that does not currently exist.

What Lehane seems to be proposing is a narrow lane of cooperation within a broader relationship defined by rivalry. The governance body would focus on safety standards and testing protocols, not on restricting capability development. It's a bet that both sides have an interest in preventing catastrophic accidents, even if they disagree on almost everything else.

Whether that bet pays off depends on whether Chinese leadership sees value in the framework. If they believe it would constrain their development while leaving American companies relatively untouched, they won't join. If they believe it would give them insight into American capabilities and a voice in global standards, they might.

What's Actually at Stake

The proposal reflects a growing concern inside the AI industry that frontier systems may soon outpace existing regulatory structures. Companies developing the most powerful models worry about cybersecurity risks, autonomous capabilities, and the possibility of hostile misuse — by nation-states, criminal organisations, or individuals.

Those concerns are not theoretical. Anthropic's Mythos model triggered White House attention after demonstrating advanced vulnerability discovery capabilities. OpenAI's own models have been used to generate exploits that were later patched. The recent Palisade Research paper showing AI agents can self-replicate with 81% success has only intensified the urgency.

The argument for international governance is straightforward: dangerous AI capabilities don't respect borders. A model trained in one country can be deployed anywhere. A safety incident in one jurisdiction can cascade globally. If the most powerful systems are going to be tested and constrained at all, the testing and constraints need to be coordinated.

The argument against is equally straightforward: international governance is slow, bureaucratic, and easily captured by the most cautious voices. If the body requires consensus among members with divergent interests, it may end up delaying beneficial applications while failing to prevent harmful ones.

The Verdict

OpenAI's proposal is significant less for its specifics than for what it signals. The company that once focused almost exclusively on capability development is now explicitly framing AI as a technology that requires nuclear-level international oversight.

That shift is partly self-interest — OpenAI would benefit from a regulatory framework that entrenches its position as a responsible actor while raising barriers for less established competitors. It's also partly genuine concern from people who have seen what these systems can do and believe the current trajectory is unsustainable without coordination.

The proposal will face serious obstacles. The Trump administration may not want mandatory oversight. China may not want to join a US-led body. Congress may not want to delegate authority to an international organisation. And the technical challenges of verifying compliance on AI systems are genuinely harder than verifying compliance on nuclear facilities.

But the underlying premise is hard to argue with: if the US and China don't find some way to coordinate on AI safety, everyone else is flying blind. The question is whether the political will exists to build the institutions before the technology makes them obsolete.

Trump and Xi are talking right now. Whether AI governance makes it onto the agenda in any meaningful way will tell us a lot about whether this proposal is the beginning of something real — or just a well-timed press release.

Sources:

  • Bloomberg. (2026-05-13). OpenAI Backs Creation of Global AI Governance Body Led by US, Would Include China.
  • Firstpost. (2026-05-14). OpenAI Proposes Global AI Watchdog With China Ahead of Trump-Xi Summit.
  • Fox Business. (2026-05-13). OpenAI Backs US-Led Global AI Governance Including China.