U.S.-China AI Treaty: A Call for Action

It’s no secret that the U.S. and China often find themselves at odds, but there’s one area where we all face risks: artificial intelligence. Recently, I, as a leading member of the House Select Committee on China, organized a historic congressional hearing to advocate for a binding, enforceable treaty on AI between the U.S. and China. Following this, I joined my fellow Democrats in urging the White House to take global steps to address AI-related dangers. Not long after, we saw Presidents Trump and Xi agree to initiate a bilateral dialogue on AI—an encouraging development that likely wouldn’t have materialized without the intensity of our push.

Yet, as important as dialogue is, it’s just the beginning. At the hearing, top experts in AI suggested that the U.S. and China form technical working groups involving leading scientists from both nations. The purpose? To create verification technologies that would help ensure neither country tries to cheat the system once a treaty is in place. This step seems, well, quite manageable and should really be initiated without delay.

Moreover, it’s crucial to start imagining what an actual treaty aimed at safeguarding humanity from AI may look like. The potential for humans to lose their grip on AI—or for skewed models to cause unpredictable consequences—is growing. If the major AI powers don’t act soon, the consequences could be dire.

There are five essential elements that should be included in a U.S.-China AI treaty. The first is a global freeze on recursive self-improvement. This concept relates to AI creating other AIs independently, which poses the greatest threat of complete loss of human oversight. Relying on AI to build its own safeguards is somewhat like trusting the fox to guard the henhouse—it just doesn’t sit right.

Next is a worldwide moratorium on superintelligence. Traditional definitions suggest superintelligence would surpass human capabilities in all areas, and frankly, that’s something neither country is truly prepared for.

Third, we need to ensure that advanced AI models are equipped with kill switches. There’s some disagreement among scientists about how effective these are, but having that option could give us a fighting chance if an AI were to turn rogue.

Fourth, it’s imperative that all AI models undergo rigorous testing in isolated environments before they’re deployed. Regardless of whether the development happens in the U.S. or China, we can’t let criminals or destabilizing AI agents use the technology for harmful purposes. The stakes are simply too high.

Lastly, and arguably most importantly, we ought to embrace President Reagan’s prudent advice: “trust, but verify.” With China, it’s essential to be cautious. Both nations must come to an agreement over verification technologies and inspection processes to guard against discrepancies that could allow one side to gain unfair advantages. Establishing a trustworthy international body for AI oversight could be vital here.

Reflecting on the Cold War, even when tensions were high, leaders like Reagan and Gorbachev recognized that nuclear conflict would lead to mutual destruction. The same can be said for an AI conflict—winning would not equate to success, but rather devastation for civilization.

Even as we challenge China on its unfair economic practices and human rights record, it’s important to unite over the need to keep human control over AI. A U.S.-China AI treaty isn’t about benefiting one nation over the other; it’s about preserving humanity as a whole. An AI disaster wouldn’t differentiate between citizens of either country, and a world where we lose our ability to steer the future is unacceptable.

Creating a verifiable U.S.-China AI treaty will take time and effort, but it’s essential to begin this journey. Additionally, we must ensure that American legislation keeps pace with the needs of humanity, preventing AI firms from prioritizing profits over societal responsibility. U.S. AI labs should also boost their cybersecurity to protect against the risk of losing vital technological advantages to China. Allowing lax cybersecurity could jeopardize our national security and weaken our position in treaty negotiations.

The potentially disastrous consequences of advanced AI technology demand a united front—it’s not about America versus China or political parties squabbling. The rise of superintelligence could be one of the defining moments for humanity, comparable to major historical shifts like the Industrial Revolution or the development of nuclear weapons. While those events transformed the landscape of society and posed existential threats, superintelligence has the potential to do both. Let’s come together to ensure humanity remains at the forefront of this unfolding narrative.