"AI could kill us all" — insider quits with warning

A senior artificial intelligence safety researcher has resigned from one of the leading AI laboratories, warning publicly that the most advanced systems now in development could, in the worst case, pose a threat to human life on a large scale.

In a resignation note published on Thursday and confirmed to reporters, the researcher said the pace of capability gains has outrun the internal review processes meant to constrain them, and that commercial pressure is shortening the time given to safety evaluations before models are released.

Why this resignation is being taken seriously

The departure is the latest in a series of exits from safety and alignment teams at frontier laboratories over the past two years. Each has followed a similar pattern: a researcher with direct access to unreleased systems argues that evaluation work is being compressed, then leaves rather than sign off on a release.

Independent researchers contacted this evening cautioned that warnings of catastrophic risk remain contested within the field. A significant body of scientists argues the nearer-term harms - automated fraud, synthetic media, labour displacement and the concentration of computing power in a handful of companies - are better evidenced than existential scenarios, and are already measurable.

What the laboratory says

The company has not disputed the resignation but says its release process includes external red-teaming and staged deployment, and that no system is released without sign-off from a safety board that operates independently of the product organisation.

What happens next

Regulators in the United Kingdom, the European Union and the United States have all begun building capability testing regimes. None currently has the statutory power to block a release outright on safety grounds. Officials in Brussels have signalled that the next review of the AI Act's general-purpose model rules will examine exactly that question.

Reported from WIRED, Axios and Reuters coverage of the resignation and from public statements by the laboratory. Verification cut-off 22:30 BST on 10 September 2026. The account of internal review processes rests on the researcher's own description and has not been independently confirmed.