A nonproliferation treaty didn’t stop the spread of nukes. It won’t rein in AI either.
The biggest supporters and critics of artificial intelligence seem to have reached a consensus: AI companies need to stop pushing the technology’s frontier until they understand how it will develop. Yet the ideas proposed by the AI safety movement cannot meaningfully reduce these risks.
Leaders of most top AI labs express deep misgivings about their responsibilities. DeepMind’s Demis Hassabis, Anthropic’s Dario Amodei, and OpenAI’s Sam Altman cosigned a 2023 statement declaring that “mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.” Though they remain uncertain about concrete solutions, the debate is urgent. Last month, thousands of AI company employees signed a letter demanding Washington support an international effort to deliberately pace automated AI development.
Sebastian Mallaby, biographer and longtime confidant of DeepMind’s Hassabis, recently warned in Foreign Affairs that “we have arrived at a perilous moment—the birth of what amounts to a new species, one that outsmarts humans.” He argued an AI equivalent to the Nuclear Nonproliferation Treaty is necessary, urging Washington negotiate joint safety principles with China and other powers.
Bernie Sanders dismissed this approach as insufficient. The same day Mallaby’s article was published, Sanders demanded OpenAI, Anthropic, and Meta pause AI development, threatening that “if you do not take appropriate action now, my colleagues and I in the U.S. Senate will.” Alongside Alexandria Ocasio-Cortez—who recently declared her previous “woke 1” movement “crazy”—Sanders is advancing a bill to outlaw new AI computing power construction in the United States and related exports. They claim this would prevent a race to eliminate hundreds of millions of jobs or build an AI capable of planetary destruction.
These concerns highlight critical issues: frontier AI models excel at manipulating code, raising profound cybersecurity implications. Donald Trump shifted this summer from a hands-off stance to limiting such releases. History shows automation’s transformative power—the Industrial Revolution reshaped society—and AI could trigger similarly sweeping change.
Unfortunately, proposed solutions are ineffective. The NPT’s record proves that technical treaties cannot prevent proliferation. When the treaty was ratified, five countries possessed nuclear weapons; since then, four more have acquired them, and many more now consider atomic capabilities. American power has historically deterred proliferation far more effectively than treaties alone—yet even with U.S. support, states like Israel built arsenals when security was uncertain, while Seoul and Tokyo actively debate similar paths.
The pausers are already too late. Mallaby warned that “follower labs” could deploy automated hacking systems using “open weight” models, which once accessed cannot be stopped. He hoped negotiations with China would prevent such tools from spreading—but Chinese hackers have already exploited them to breach Taiwanese government websites.
Washington lacks the authority to provide guarantees for an NPT-style treaty to work. Cyberattacks are commonplace; no rational leader would relinquish advanced cybersecurity tools in hopes of punishing attackers. AOC and Sanders’ approach is even more misguided. When Israeli cybersecurity experts detected the Taiwan attack, they stated: “the basic assumption of every government around the globe” should be perpetual cyber vulnerability. The left’s solution—letting other nations surpass U.S. companies—is reckless when a country under active attack cannot afford to lag behind.
Unilateral disarmament does not work, especially for a nation already under siege.