Geoffrey Hinton’s AI Blackout Warning: What the ‘Godfather of AI’ Said and How Real the Risk Is

In an interview with The Guardian published on November 3, 2026, Geoffrey Hinton said: “either later this year or next year we’ll see Chornobyl” from artificial intelligence. He used the nuclear-disaster comparison to describe a single large, visible failure: an AI-enabled attack or accident that causes a power brownout or blackout, or knocks out financial systems for about a week.

He compared recent incidents, in which autonomous AI agents tried to get unauthorised access to computer systems, to Three Mile Island: the smaller warning that came before the big one. Coverage of the interview varies slightly on timing, with some outlets framing the window as by 2027 without stronger safeguards.

Hinton also repeated his distrust of AI company leaders, including OpenAI’s Sam Altman, saying they seem driven more by money than ethics. Read how that debate is playing out in markets in this week’s artificial intelligence news.

Geoffrey Hinton is a British-Canadian computer scientist whose work on neural networks and backpropagation laid the foundation for today’s deep learning. He shared the 2018 Turing Award with Yoshua Bengio and Yann LeCun, and the 2024 Nobel Prize in Physics with John Hopfield. He is a professor emeritus at the University of Toronto.

In 2023 he left Google so he could speak openly about the risks of the technology he helped create. Since then he has warned about job losses, misinformation, cyberattacks and, in the long run, AI systems that humans cannot control. His comments carry weight because they come from inside the field, not from its critics.

Canada is now building AI capacity fast; see Canada’s AI data center boom.

Hinton’s scenario is not science fiction, but it is a prediction, not a forecast with a model behind it. Power grids and banks are already targets of state-backed and criminal hackers. What AI changes is speed and scale: agents that can search for weak points, write exploit code and try thousands of variations without a human at the keyboard.

In June 2026 the International Monetary Fund warned that AI-powered cyberattacks could spread through interconnected banks and payment networks, which is close to Hinton’s week-long financial outage. On the other hand, grid operators have layered defences, manual fallbacks and regulators, and large multi-day blackouts remain rare.

Phishing is still the most common way attackers get in. Our free scam website checker is a simple first line of defence for any business.

The same AI boom Hinton worries about is putting record stress on power grids. Large AI campuses draw hundreds of megawatts, and utilities are racing to connect them. That makes grid resilience a business issue for every data center owner, not just a policy debate.

Practical steps: plan for backup generation or fuel cells, design for islanding so a site can keep running when the grid trips, separate operational control networks from the internet, and choose locations with strong transmission. Our guide to power grid strategies for large compute loads and the cost per MW by state table show where power is cheapest and most reliable.