Anthropic safety lead puts >10% odds on AI killing all humans as researcher quits
nordicinst · x · 2026-09-09
Per The Verge, Anthropic safety researcher Jacob Coxon announced his departure, accusing Anthropic and OpenAI of "racing straight to self-improving superintelligence and gambling with our lives," even as builders "earnestly believe it could kill us all by the end of the decade."
Hours after the resignation, Anthropic's safety lead publicly put the odds of AI "killing all humans" by 2030 at more than 1 in 10.
- Coxon previously trained systems at both Anthropic and OpenAI
- Insiders have long warned about recursive self-improvement spiraling out of control; much of today's code is already AI-written
- The departing researcher says AI companies are chasing uncontrollable superhuman systems
More from AGI Musings
- AI engineer quits citing 10% extinction risk; peers agree but the race won't stop — IgorCarron · 2026-09-09
- US and UK bills to ban ASI gain backing from Nobel laureate Geoffrey Hinton — Puzzled-Ad-6854 · 2026-09-09
- As AI makes execution cheap, good ideas become the real currency — StefanoGogioso · 2026-09-09
- Politician letters UN, OECD and G7 urging a treaty for safe superintelligence development — S_OhEigeartaigh · 2026-09-09
- Eric Weinstein calls AI's trophy-value capture from researchers a market failure — hugobowne · 2026-09-09
- Fans decode Iron Man's best Easter egg: JARVIS as Tony Stark's failsafe — aakashgupta · 2026-09-09