Geoffrey Hinton said his personal estimate of the chance that AI could destroy humanity is greater than 1%. In an interview published by EL PAÍS on October 21, 2026, he also proposed requiring statistical evidence of chatbot safety before launch, reviewed by a government agency.
Hinton’s personal estimate of AI risk
Hinton described the chance of AI destroying humanity as significant and greater than 1%. The figure is his own judgment, not a consensus estimate.
AI-risk percentages can refer to different outcomes and time frames. NeoTeo previously explored those differences in an analysis of AI catastrophe estimates.
Two proposals for AI safety
Hinton’s regulatory proposal would require a chatbot’s safety to be demonstrated statistically to a government agency before the chatbot launches. It assigns a public agency a role in reviewing safety evidence before release.
In separate reporting, Fox Business said Hinton had also advocated giving advanced AI “maternal instincts,” describing “AI mothers” rather than assistants. That is a design idea about how a system might behave; the government review proposal concerns oversight before launch.
