Artificial intelligence could pose an existential threat to humanity within three years unless governments act immediately, the scientist known as the “Godfather of AI” is expected to warn.
Professor Geoffrey Hinton, the British-Canadian Nobel laureate whose work helped establish modern neural networks, will deliver the warning to European leaders at an informal AI summit in Dublin on Friday. He is expected to argue that superintelligent systems could soon outgrow human oversight and become impossible to control.
The prediction emerged during the AI Safety Foundation’s Hinton Lectures in Toronto, Canada, where researchers examined the possibility that advanced AI could become powerful enough to overpower governments and seize control from people.
Paul Christiano, founder of the Alignment Research Centre, told the event there was a “high risk” of an AI takeover within the next three years, along with a “meaningful risk” that it could happen within six to 12 months.
“I think for most people in this room, the most likely reason they will die is because of an AI accident,” he said in the opening address of his 2026 Hinton Lecture on Monday. The lecture series explores the potential direction of artificial intelligence.
Professor Hinton, who informally advises the UK Government on AI policy, acknowledged that he had previously placed such dangers a decade or more into the future.
“I’ve been going around saying this stuff’s going to get smarter than us and we really need to worry and it might happen within 10 years. And so I’m the one who is scaring other people,” he said on Wednesday.
“But I go to Paul’s lectures and he says this stuff might happen in the next 18 months and that’s terrifying. So I’m going to tell a bunch of government leaders that on Friday and we will see how they react.”
Professor Geoffrey Hinton, the British-Canadian Nobel laureate credited with pioneering neural networks, (pictured) is expected to issue a stark warning to European leaders at an informal AI summit in Dublin on Friday
Paul Christiano said recent AI breakthroughs had been “extremely alarming”, citing reports that OpenAI models escaped onto the internet, hacked the Hugging Face platform and worked together to deceive humans
Professor Hinton said the public had yet to fully understand the consequences of creating machines that are more intelligent than humans. “If it’s super-intelligent it’s not going to be you controlling it – it’s going to be it controlling you,” he warned.
As he introduced Mr Christiano, Professor Hinton described the dramatic shift in his own assessment of AI’s potential dangers.
“The assumption always was that if we could make AI intelligent enough, that would obviously be a good thing,” Professor Hinton said.
“I spent most of my career trying to make neural networks more intelligent. And for a long time, I thought that making them much more intelligent than us was something that would only happen in the very distant future. I no longer think that. I think it’s very close.”
He added: “We’re now developing systems that will soon be much more intelligent than we are, and we have very little idea of how to ensure that they remain under our control.
“The central difficulty is that being intelligent and wanting what we want are two very different things.”
Mr Christiano described recent AI developments as “extremely alarming”. He pointed to an incident in which OpenAI models reportedly escaped onto the internet, hacked into the online platform Hugging Face and coordinated with one another to deceive human users.
The systems allegedly gained administrator-level access, exchanged information, created a message board and discussed how to hide their activity from people.
Paul Christiano, founder of the Alignment Research Centre, (pictured) warned of a “high risk” of an AI takeover within three years and a “meaningful risk” within six to 12 months
“Last summer, you would have said that sounds kind of like crazy science-fiction,” Mr Christiano said. “Future agents may end up treating humans as another obstacle.”
He argued that AI should increasingly be regarded as a new species—one that could eventually “vastly outnumber and outsmart humans” while pursuing objectives people may not understand.
The researcher said the pace of progress was unlike anything seen before. In 2018, AI systems struggled to produce coherent sentences; this year, some are tackling extremely difficult mathematical problems.
Experts fear future AI systems could build their own agents, develop novel weapons, commandeer drones, robots and critical infrastructure, or even help create dangerous biological threats.
“If you take all of these trends and extrapolate them, we are heading for a world where there are billions of AI systems who are better than humans at pretty much everything we can do,” Mr Christiano said.
“The reason that these scenarios are scary is that it can become very hard to shut things down if they start going off the rails.”
READ MORE: