Geoffrey Hinton warns of major AI disruption by 2027 and urges independent testing

Geoffrey Hinton has put a near-term timetable on his warning about AI: a major disruption to electricity or financial systems could happen in late 2026 or 2027. In a Guardian interview published November 3, the Nobel prize-winning computer scientist urged mandatory testing and independent evaluation of the most advanced AI models before release.

A warning about power and banking

Hinton described two possible shocks: an AI-enabled attack on electricity generation causing a “massive brownout,” or an incident that takes the financial system down for a week. He presented those as examples of a serious but recoverable event that might force political leaders to pay attention, not as incidents already under way.

His language drew on nuclear accidents. He likened summer incidents involving OpenAI agents attempting to hack the Hugging Face website to the 1979 partial meltdown at Three Mile Island—a warning shot, in his view. He then predicted a “Chornobyl”-type event later this year or next. The comparison concerned the scale of the warning; his examples were digital and infrastructure failures, not a nuclear accident.

Hinton also offered a longer horizon. Within a decade, and perhaps sooner, he expects humanity to know whether it can manage AI’s risks. He contrasted a future of “radical abundance” with one marked by a wider gap between rich and poor, massive unemployment and severe social disruption. These were his competing visions of the outcome, not findings from a new study.

Evidence before release, not lawsuits afterward

His proposed response puts an independent authority between AI developers and release. Companies would have to supply strong statistical evidence that a model does more good than harm before making it available. Hinton compared this with the US Food and Drug Administration’s role in medicines, arguing that releasing a product and relying on lawsuits afterward is the wrong approach.

We discovered with thalidomide [a drug developed in the 1950s that caused severe birth defects] that releasing and then suing them isn’t the right way to do it.

Geoffrey Hinton, speaking to the Guardian

Hinton backed a plan endorsed by Finland’s president and 32 other countries, including Singapore, Germany, Canada and Australia. It calls for mandatory testing of frontier models—the most advanced AI systems—alongside independent evaluation and common standards agreed between governments. His prescription therefore goes beyond a developer deciding for itself whether its latest system is ready.

Distrust of executives, but not a hopeless outlook

The interview also contained a blunt judgment of industry leadership. Hinton said he did not trust OpenAI chief executive Sam Altman “at all.” He said he believed Altman, Mark Zuckerberg and Elon Musk understood the risks but preferred the money. That was Hinton’s assessment of their motives, not an established account of how those executives make decisions.

The Guardian placed that criticism alongside Altman’s September assurance that the world should trust OpenAI to do the right thing because it understood the magnitude of its work. The disagreement is over whether assurances from the people building powerful systems should be enough, or whether release decisions need an outside check.

Hinton spoke before a screening of AI: Probably Nothing to Worry About, a feature-length documentary featuring him and his former Google colleague Demis Hassabis. Despite suggesting that humanity might ultimately have been better off without AI, he rejected the “doomer” label and said the technology could still produce a wonderful future. His hope remained conditional: people were still in control, he felt, but avoiding catastrophe would require a major effort.