Geoffrey Hinton has spent the past year telling anyone who will listen that AI could slip out of human control. On Friday, the people he told was not a podcast audience or a Senate committee — it was the assembled leadership of the European Union, meeting in Dublin for their first-ever dedicated discussion on artificial intelligence.
The informal summit, arranged by European Council President António Costa, split its AI session in two halves that map the bloc’s anxieties almost perfectly. The first covered competitiveness, addressed by Arthur Mensch, CEO of France’s Mistral AI, who urged the leaders to “act now” to keep pace with the United States and China. “Europe has all the technology and the talent to lead the AI race,” he said, according to Deutsche Welle. “What it needs is speed, financing and a strong coordination between member states.”
The second half belonged to Hinton. The 78-year-old Nobel laureate, who informally advises the UK Government, has been cranking his own risk estimate upward in public: he has floated a 10 to 20 per cent chance that AI could cause human extinction by escaping human control, and this week told an audience that “for most people in this room, the most likely reason they will die is because of an AI accident” — a line he previously dated “maybe 10 years out” but now treats as much closer. As reported by Metro, Hinton said ahead of theDublin session: “I’m going to tell a bunch of government leaders that on Friday and we will see how they react.”
His sharpened timeline owes a debt to Paul Christiano, the alignment researcher who used the opening lecture of the AI Safety Foundation’s 2026 Hinton Lectures in Toronto to warn of a “high risk” of AI takeover within roughly three years. “I go to Paul’s lectures and he says this stuff might happen in the next 18 months and that’s terrifying,” Hinton said at the same event, per Metro. Christiano, notably, also sits on OpenAI’s foundation board — a detail that makes his warnings land differently than they would from a detached academic.
A summit almost swallowed by its own agenda
The irony of the day: Europe’s leaders gathered to discuss the technology that might redefine their economies, and spent most of it arguing about Ukraine. The question of who should represent the EU in peace talks — contested between European Council President Costa, foreign affairs chief Kaja Kallas and the Germany-France-UK “E3” format — hijacked the proceedings, as Euronews chronicled. The AI discussion survived, but only in the informal, no-communiqué format its organisers chose explicitly because it “gets overcrowded by current events” less easily.
Kallas supplied the most concrete risk framing of the day, pointing to AI-guided drones in Ukraine: “We can all already see this in the battlefield in Ukraine, where you have drones equipped by artificial intelligence to really train to find anybody to kill,” she said, per DW. And in a letter ahead of the summit, the governments of Estonia and Finland called for EU-level arrangements for responding to serious AI incidents — a striking admission that some incidents are now considered a matter of when, not if.
European Parliament President Roberta Metsola pushed back on the doom-adjacent framing: “I am frustrated with those who say Europe has somehow lost the AI race.” She has a point about the framing, if not the scoreboard — the frontier labs building the systems Hinton worries about are overwhelmingly American and Chinese, and Mistral is Europe’s only meaningful counterweight. The EU is now advancing rules to steer public contracts in sensitive use cases toward European companies, an industrial-policy answer to a capability gap.
What it means for everyone else
There is a telling asymmetry in how the two guest speakers framed the same hour. Mensch asked the room for money and speed; Hinton asked it for caution. Both positions imply the same structural fact — the decisions that determine how capable AI systems become, and how controlled they stay, are being made in a handful of boardrooms and capitals, by a small set of people, very fast. Whether Europe’s contribution turns out to be “sovereign compute” or “the precautionary counterweight” may matter less than that it keeps a seat at the table at all.
For New Zealand and other small economies paying attention, the Dublin session is a preview of the fork every government now faces: chase the productivity gains and accept the exposure, or regulate early and watch frontier investment route around you. Hinton’s own government relationship — heinformally advises the UK, which has consistently declined to follow the EU’s AI Act — shows even close allied governments are hedging between the two positions, as we noted when the EU’s enforcement era began.
The most concrete outcomes from Friday: none. That is the nature of an informal summit, and by design. But the symbolism is not nothing. Hinton has testified to the US Senate — we covered his “maybe a year” warning to Congress in September — and advised the UK. Addressing the full European Council is the last major Western capital-level audience he had not reached. After Friday, the argument that nobody saw this coming has one fewer place to hide.
Hinton’s warning tour has company: 415 scientists, including several of his fellow Turing Award winners, signed an open letter this spring demanding airgaps and a global watchdog for frontier systems ([/415-scientists-ai-containment-open-letter-2026/]), and nonprofits are currently in a California courtroom asking a judge to freeze frontier AI development outright. The gap between those demands and what any government has actually done remains the widest it has ever been — and Hinton just made sure Europe’s leaders couldn’t say they hadn’t heard it firsthand.
