REP RO KHANNA: US and China can't afford an AI arms race with humanity at stake
America and China do not agree on much. But when it comes to AI, we all lose if we lose control.
Last month, as the ranking member of the House Select Committee on China, I convened the first-ever congressional hearing calling for a binding, enforceable, and verifiable U.S.-China AI treaty. I also led every Democrat on the Committee in sending the White House a letter demanding global action to tackle AI risks. Within days, President Trump and General Secretary Xi announced that they would establish a bilateral dialogue on AI. This is an important first step that would not have happened without congressional pressure.
But this moment demands more than discussion. At my hearing, some of the world's leading experts on AI called for the U.S. and China to form joint technical working groups with both countries' top scientists. Their mission would be to develop mutually trusted verification technologies that prevent either side from cheating once we sign a U.S.-China AI treaty. This is an easy second step we should take today.
ANTHROPIC CEO LIKENS AI FIGHT WITH CHINA TO COLD WAR, SEEKS 'DISARMAMENT NEGOTIATIONS'
Critically, we also need to begin envisioning what an actual treaty to save humanity from unchecked AI would look like. The risk that humans will lose control of AI or that misaligned models could unleash biological havoc is increasing every day, and if the world's AI superpowers do not act soon, it may be too late.
To address these risks, there are five key pillars of a U.S.-China AI treaty. First, we need a worldwide pause on recursive self-improvement. This is when AI develops AI without human intervention, and it's the most likely path to humanity losing control of this technology entirely. Asking AI to design its own safeguards is like asking the fox to guard the henhouse.
MIKE JOHNSON REJECTS AI MORATORIUM OVER CHINA THREAT AS CONSERVATIVE ACTIVIST DEMANDS GUARDRAILS
Second, we should impose a global pause on superintelligence. As it has been traditionally defined, superintelligence is AI that exceeds human capabilities across all domains. This could unleash Pandora's box, and neither America nor China is ready for it.
Third, we must have kill switches embedded in all advanced models. While scientists disagree about the efficacy of kill switches, they would at least give us a fighting chance if rogue models decided to turn against humanity.
AI KILL SWITCH BILL COULD SHUT DOWN ROGUE MODELS
Fourth, every AI model should be tested in an air-gapped sandbox prior to release to prevent catastrophic misuse. Whether it's an American model or a Chinese model, criminals, terrorists and even misaligned AI agents shouldn't be able to unleash economically devastating cyberattacks because a frontier lab decided to release a beta version.
And fifth, but perhaps most importantly, we need to heed President Reagan's wise words and "trust, but verify." Especially with China, we cannot simply take the CCP at its word. Both governments should agree on verification technologies and inspection regimes that ensure Beijing cannot simply cheat, allowing its labs to race ahead while American labs lag behind. We will also need to build institutions that both sides in a fierce competition can trust, such as an International AI Inspection Agency.
BEYOND KILLER ROBOTS: CHINA'S REAL AI THREAT TO AMERICA IS ALREADY HERE, EXPERTS WARN
During the Cold War, even as America and the USSR fought fiercely, President Reagan and Soviet leader Mikhail Gorbachev recognized that nuclear war would have left both nations worse off. Winning an AI war is like winning a nuclear war. It is catastrophic for civilization.
In that same spirit, even as we continue to hold China accountable for its predatory economic practices that harm American workers and its human rights abuses, both countries should be able to rally around the common cause of keeping humans in control of AI. A U.S.-China AI treaty advantages neither America nor China, but rather all of humanity. A global pandemic unleashed by rogue AI wouldn't distinguish between American and Chinese citizens, and neither government benefits if mankind ceases to call the shots.
CLICK HERE FOR MORE FOX NEWS OPINION
A verifiable U.S.-China AI treaty will not happen overnight, but if we do not begin the hard work of negotiating such an agreement soon, humanity will pay a steep price. In the meantime, in addition to forming joint technical working groups with China, we need to get our own house in order.
For Congress, this means passing laws that prevent AI companies from putting profit over the collective fate of humanity. For American AI labs, this means investing in safety and strengthening their cybersecurity to prevent China from stealing our edge. If lax cybersecurity allows China to obtain our most advanced model weights illicitly, it not only harms our national security, but it also undermines our strongest leverage in pursuit of a U.S.-China AI treaty.
When it comes to catastrophic AI risks, it cannot be America versus China or Republicans versus Democrats. The arrival of superintelligence arguably marks a more consequential moment for our species than even the dawn of the Industrial Revolution or the development of the atomic bomb. While the former transformed society and the latter risked its very destruction, the most advanced forms of AI threaten to do both. Let's come together and keep humanity in the driver's seat.