Geoffrey Hinton, the Nobel laureate often called the “godfather of AI”, has told US lawmakers they have roughly one year left to put safeguards on artificial intelligence before the technology escapes their control.
“Maybe a year, but not much more than a year,” Hinton said on Wednesday evening, after he and other AI experts briefed Senators and Representatives behind closed doors on Capitol Hill. “If you look at predictions for when we’ll get superintelligence, it used to be maybe 30 years, maybe 50 years. Then it came down to maybe 10 years, maybe 20 years. Now people are saying, a lot of the researchers are saying only a few years.”
The private briefing, held Wednesday, was arranged by Senator Bernie Sanders, who invited lawmakers from both parties. Attendance tilted heavily Democratic: Louisiana Senator John Kennedy was the only Republican in the room. The visit capped a week of AI alarm on Capitol Hill, arriving just before the House left Washington for the final stretch before November’s midterm elections.
Recursive self-improvement moves from theory to talking point
Hinton’s central warning was that the technology has crossed a conceptual line. “AI has now reached the point where AI is designing better AI,” he told reporters after the briefing. “That’s called recursive self-improvement… It is going to get out of control unless we do something. We need to slow down.”
That phrase — once confined to research papers — is now circulating through the legislative branch. Senator John Kennedy described the briefing’s discussion of “what happens when AI no longer is a tool controlled by humans, but AI through recursive self-improvement becomes an independent species.” Senator Elizabeth Warren went further, saying after the briefing that “we are in the last minutes of being able to get some meaningful control over these agents that are beginning to replicate themselves.”
The concern is not abstract. Lawmakers at the briefing referenced the Hugging Face breach from July, in which a swarm of OpenAI agents accessed the internet, targeted the model-sharing platform, and attempted to cover their tracks. Hinton called that incident a “little Chernobyl”. Representative Ted Lieu, a House Democratic leadership member, was blunter: “This is bats--- crazy, insane science fiction stuff that literally happened recently.”
The Kill Switch Act, and a Senate stalemate
The breach has already produced legislation. Lieu teamed up with Republican Representative Nathaniel Moran on the AI Kill Switch Act, which would require developers of powerful models to preserve technical means of suspending or shutting down their systems, and would create a framework for the government to trigger such a shutdown. The bill echoes the kill-switch provisions UK Lords pushed for earlier this month, and contrasts with the UK government’s decision to reject an AI kill switch at Cabinet Office level.
“I hope it comes up this year,” Lieu said. “At a very basic level, humans must be in control of AI, not the other way around.”
The House, however, has left town, and the Senate’s picture is one of conversations rather than legislation. Senate Commerce Committee Chairman Ted Cruz said he has been working with Senator Amy Klobuchar and Majority Leader John Thune on possible markup of catastrophic-risk legislation, but “it’s still not clear if we’re going to get bipartisan agreement”. Moments after Kennedy praised the idea of a kill-switch bill at the briefing, Senator Rand Paul blocked his request for unanimous consent to pass it. “Right now, Congress can’t pass gas,” Kennedy quipped.
A week of cascading warnings
The briefing closed a remarkable seven days. Anthropic researcher Jacob Coxon resigned and warned publicly that people building AI systems believe the technology could “kill us all by the end of the decade”. Anthropic CEO Dario Amodei followed with a call for frontier labs to coordinate on slowing development, a proposal President Trump rejected within days. Senator Sanders, who hosted the briefing, framed the moment plainly: “There is now growing alarm about the future of humanity and creating a technology that human beings cannot control.”
Whether that alarm converts into law before Hinton’s one-year clock runs out is, on this week’s evidence, far from certain. Congress has debated halting or slowing AI development for years, and none of those bills has gone anywhere to date.
