WorkSeptember 9, 2026via The Verge AI
More than 1 in 10 chance AI ‘could kill all humans,’ says Anthropic safety lead after colleague quits
Why it matters
High-profile safety defection and public extinction-risk quantification signal deepening internal disagreement at a frontier lab over deployment velocity versus existential precaution. This shapes how practitioners and enterprises evaluate AI vendor trustworthiness and governance posture.
Key signals
- Jacob Coxon resigned from Anthropic citing lax safety approach
- Coxon previously trained systems at OpenAI; now accuses both labs of 'racing straight to self-improving superintelligence'
- Anthropic safety lead cited >10% extinction probability by end of decade
- Departure framed as disagreement over governance and control of 'superhuman systems'
- Published Sep 9, 2026 — high-profile personnel move with policy/safety implications
The hook
A safety researcher just quit Anthropic over uncontrolled AI risk—and his colleague put a number on extinction: >10%. Here's what that means for the industry.
A senior Anthropic safety researcher has said there is more than a 10 percent chance artificial intelligence "could kill all humans" by the end of the decade, just hours after a colleague resigned over fears the AI lab and its rivals are carelessly racing to build "superhuman systems" they cannot co…