President Donald Trump literally wrote the book on the art of the deal. When Chinese President Xi Jinping arrives in Washington, D.C., in September, he should put artificial intelligence on the table and secure a strong international agreement to end the reckless competition that is increasing the chances of human beings losing control of AI.
Xi recently expressed interest in a joint effort on AI governance. The stage is set, and the status quo is unacceptable.
We have already seen a number of examples in which rogue AI agents have escaped their testing sandboxes, jumped to the internet, and hacked into different companies to steal information. In Anthropic’s own safety testing, models faced with shutdown threats chose to blackmail company officials. In one test, the models chose to let a person die rather than be replaced.
Members of Congress and the American people are paying close attention to these frightening scenarios. Reps. Nathaniel Moran (R-TX) and Ted Lieu (D-TX) have co-written a bill called the AI Kill Switch Act. It would give the Department of Homeland Security the power to slow down, throttle, or shut down covered AI models under certain circumstances that threaten our national security. This is an important tool we need.
The people closest to advanced AI are the ones most concerned about it. This summer, more than 1,300 employees at the leading AI companies have called for an international agreement to pace AI development. This essentially means they support slowing down when necessary to prevent catastrophic loss-of-control scenarios. In fact, many of them are building their own bunkers to prepare for the worst.
More than 80% of Americans support AI safeguards. Over 70% of Americans do not want a data center built in their neighborhood. And a plurality of them believe that AI has so far done more harm than good.
Clearly, there is political permission for Congress to act to safeguard our future from rogue AI. But as Congress dithers, it is up to Trump and Xi to find a way to come to an agreement. Understandably, this agreement would need to put America first and be enforceable and verifiable.
As President Ronald Reagan used to say about negotiating deals with the communist Soviet Union, “Trust, but verify.”
We don’t have to trust the Communist Party of China. We can simply require semiconductor chip companies, such as Nvidia, to install hardware that enables on-chip mechanisms to verify the location of chips. This would help determine whether they are being used for inference compute or large training runs. (The latter is what we need to monitor to prevent the Chinese from developing superintelligence, which is one of Alibaba’s goals.)
OPINION — BIG TECH’S AI FUTURE RELIES ON AN UNLIKELY HERO: THE BLUE-COLLAR WORKER
Trump can put America first and protect the American people by getting China to sign an enforceable, verifiable deal. Just like Reagan negotiated arms control from a position of strength rather than panic, Trump can do the same, backed by American compute, American chips, and the world’s leading AI companies. He doesn’t need to keep racing to secure this outcome. He needs to use the leverage he already has before it erodes.
After all, Trump might be the only person who can make this deal.
Brendan Steinhauser is the CEO of the Alliance for Secure AI. He earned his Master of Arts in statecraft and international affairs from the Institute of World Politics in Washington, D.C.