The 'Godfather of AI' says it's not 'unreasonable' to believe there's a 10% chance the technology will end humanity

The "Godfather of AI" says it's difficult to estimate the chance AI will wipe us all out, but it's not zero.

"A 10% chance seems not an unreasonable estimate to me, but nobody really knows how to give a sensible estimate," Geoffrey Hinton said during an interview Wednesday on the BBC's "Newsnight."

Newsnight anchor Victoria Derbyshire expressed amazement that Hinton largely agreed with the assessment of a top Anthropic safety researcher who said that AI there is ">10%" chance that AI could kill all humans "within the next decade."

"Wow," Derbyshire said. "Oh, my God."

Hinton said it's hard to estimate the exact number because society has never seen a technology that "could soon be smarter than us."

"It would be very foolish to say there was like a 1% chance," he said. "Nobody knows how to estimate it."

What is clear, the Nobel laureate said, is that the window when such a catastrophe could occur has shrunk.

"I used to think like 30 to 50 years, then I reduced it to 10 to 20 years," he said. "Now, I think maybe 10 years, maybe less. And other people think much sooner than that."

As to how it could happen, Hinton mentioned the risk of everything from cyberattacks to AI developing a deadly virus.

While Hinton did not say why that timeline has compressed, leading AI companies like Anthropic and OpenAI have poured billions into scaling up their models, and their recent releases have shown significant improvements on models that are just months old.

The end of the world as we know it?

Hinton's comments come after Anthropic researcher Jacob Coxon announced his resignation from the company ahead of a hotly anticipated IPO. Coxon, who previously worked for OpenAI, said the AI giants are "gambling with our lives."

"The people building AI earnestly believe that it could kill us all by the end of the decade," Coxon wrote on X. "This is not a marketing stunt. If anything, many executives and senior researchers will couch their phrasing in the press to sound sensible - but I hear the same people express fear privately."

In response to Coxon's comments, Evan Hubinger, alignment science lead at Anthropic, wrote on X, "Jacob is correct here—we really do earnestly believe AI could kill all humans!"

"I personally think it is >10% within the next decade," Hubinger wrote. "I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to."

Hubinger later added that Anthropic has said the risk from current models is low. What's at issue, he said, is "superintelligence arising from recursive self-improvement, as we have said is happening faster than we thought."

Recursive self-improvement, or RSI, is the theoretical process by which AI researchers think AI models could improve themselves, potentially accelerating the pace of advancement as the technology would be capable of designing and training its own successors.

Hinton said humanity could go down the path toward its destruction or figure out "how to deal with the dangers" AI poses while at the same time realizing its promises. He renewed his call for global regulation, noting that while countries may not agree on everything, they share a desire to preserve humanity.

"For things like preventing AI from taking over from people, the interests of all the countries are aligned," he said. "So they will eventually collaborate on that. The question is, will they do it in time?"