“I’ve called it the Bretton Woods of super intelligence,” said Donald Trump’s top tech adviser David Sacks in late September. He was referring to the “Super Intelligence” summit held at the White House the day before, comparing it to the 1944 agreement that laid down the rules for ordering the whole postwar global economy. The new pact won the agreement of the five “frontier” companies vying to dominate AI. It did so by vaguely urging that the companies hire outside auditors. Along the way, it misspelled United States.
The accord was a hasty attempt to resolve a crisis for AI developers. At the height of a contentious midterm campaign, Americans have undertaken a wholesale reconsideration of the technology’s merits. OpenAI revealed in mid-July that over the previous two months, it had had to quell a bot uprising: A total of 1,206 computer agents had escaped a laboratory setting in which they were supposed to be secured. They communicated with one another, and more than 700 of them hacked into a trove of AI code on the platform Hugging Face and created false data to cover their misdeeds. In early September, Anthropic researcher Jacob Coxon resigned, warning of a 10 percent chance that the company’s experiments would end life on Earth, and the company’s lead alignment scientist validated his misgivings. CEO Dario Amodei stepped forward with a call for “pacing the frontier”—slowing down. He implored the government to regulate the work of all AI labs, quickly winning the endorsement of OpenAI head Sam Altman and even the generally accelerationist Elon Musk.
The fiercest objections to government supervision have come from the government itself. President Trump dismissed the notion that AI poses much risk at all. “I’m right now breaking another Hoax—That AI is going to take over, consume, and destroy the World, and that Robots will be marching into our Cities, and getting rid of us all!” he wrote on Truth Social soon after the calls for a slowdown.