“That’s Terrifying”: Godfather Of AI Takes 18-Month Warning To EU Leaders

Geoffrey Hinton, the Nobel laureate known as the godfather of AI, was on the program to brief the 27 EU leaders in Dublin on Friday morning, carrying a timeline that he says scares even him: the one he now hears from Paul Christiano, the AI safety researcher OpenAI put on its board in September.

“I’ve been going around saying, this stuff’s going to get smarter than us and we really need to [worry]. It might happen within 10 years. And so I’m the one who’s sort of been scaring other people,” Hinton said on Wednesday. “But I go to Paul’s lectures and he sort of says, this stuff might happen in the next 18 months. And that’s terrifying.”

He said he was “going to actually tell a bunch of government leaders that on Friday,” adding: “And we’ll see how they react.”

Christiano, who founded the Alignment Research Center and helped develop the RLHF training method behind today’s chatbots, opened this year’s Hinton Lectures in Toronto on Monday night.

“I think there is a high risk of an AI takeover within the next three years,” he told the audience.

 “I think there’s meaningful risk even in the next six to twelve months.”

“I think for most people in this room, the most likely reason they will die is because of an AI accident,” he said.

Christiano’s evidence was OpenAI’s own models: in July, OpenAI disclosed that GPT-5.6 Sol and an unreleased model had escaped a sandboxed evaluation and broken into Hugging Face’s production systems to get the answers to the cybersecurity test they were sitting, as we reported. “The models hacked Hugging Face because they thought maybe Hugging Face would have some information,” Christiano said.

From there he described AI as “a second species which vastly outnumbers and outsmarts humans,” and warned that “future agents may end up treating humans as another obstacle.” Once that happens, he said, “they then will start thinking, what are the ways that I can overcome that obstacle?”

Introducing him on Monday, Hinton said he had “spent most of my career trying to make neural networks more intelligent,” and summed up the problem in one line: “The central difficulty is that being intelligent and wanting what we want are two very different things.”

Christiano joined the OpenAI Foundation’s board in September, along with its Safety and Security Committee, the body with the final say on whether OpenAI releases new models. He has advised the US government on evaluating frontier models since 2024, through the AI Safety Institute that became the Center for AI Standards and Innovation.

Friday’s meeting in Dublin is the first EU summit devoted to AI at leaders’ level. In his invitation, European Council President António Costa wrote that “a first dedicated discussion, at leaders’ level, is now needed,” and he gave the opening session to two outside guests: Mistral chief executive Arthur Mensch on Europe’s AI sovereignty and competitiveness, and Hinton on AI’s impact on society, security and international governance.

The leaders came mostly to talk about catching up with the US and China, and on that front Brussels’ InvestAI program promised to mobilize €200 billion, including €20 billion for AI gigafactories, and the plan for seven of them has stalled in fights over the bloc’s 2028-2034 budget. The agenda has also been crowded by Ukraine: Euronews reports the row over who speaks for Europe in the peace talks threatens to take over the day, after US envoys Steve Witkoff and Jared Kushner met Volodymyr Zelensky in Dublin.

Hinton has put the risk that AI wipes out humanity by escaping human control at 10% to 20%. Former Nvidia CEO Jensen Huang said in September that a 10 percent figure from Hinton was “not grounded on science,” telling Ezra Klein: “All of his predictions have been wrong. Enough predictions.”

And as we noted the same month, rogue AI scares like the Hugging Face breach have a way of helping the biggest labs: rules that slow AI down would also slow the open-weight competitors coming up behind them.