Computer scientist Geoffrey Hinton has warned that artificial intelligence could soon contribute to a major disruption of critical systems, while emphasizing that the outcome is a risk assessment rather than a prediction of certainty. In an interview with the Guardian, Hinton cited a temporary failure of electricity generation or financial infrastructure as examples of an incident that could force political leaders to treat AI safety more urgently.
Hinton, a Nobel Prize winner whose research helped establish modern neural networks, said such an event might occur this year or next. He compared the possibility to an early industrial warning that changes public and government behavior, rather than to a nuclear accident itself. His examples included an AI-enabled attack causing a large brownout or taking financial systems offline for a week.
The researcher’s warning combines near-term security concerns with longer-term social risks. He said the next decade may show whether societies can use AI to create broad material benefits or instead experience greater inequality, unemployment and cultural disruption. He also said it was possible humanity would ultimately conclude that developing AI had been a mistake, although he continued to describe a positive future as achievable.
Hinton left Google in 2023 so he could discuss AI risks more freely. He has since pressed for stronger oversight of increasingly capable models and criticized reliance on voluntary assurances from leading technology companies. In the interview, he supported mandatory testing of frontier systems, independent evaluation and common standards between governments.
His preferred model resembles pre-market regulation in medicine: developers would need to provide statistical evidence to an independent authority that a system’s expected benefits outweigh its harms before release. That approach would move safety checks ahead of deployment instead of depending primarily on litigation or remediation after damage occurs. A group of countries has endorsed elements of stronger frontier-model testing, though governments remain divided over how much regulation is compatible with technological competition.
The comments also accompanied a documentary featuring Hinton and Google DeepMind chief Demis Hassabis. Hinton said his public reputation gives him a duty to speak, but rejected comparisons that cast him as the sole creator of the technology, noting the large community of researchers involved in neural-network development.
Despite the severity of his warnings, Hinton said he believed humans were still in control and might mobilize enough effort to manage the danger. His claims about the timing and form of a future incident remain informed speculation. The central policy message is more concrete: powerful systems should face independent evidence-based assessment before their release, because waiting for a disruptive failure could leave governments responding only after preventable harm.
