Anthropic Alignment Lead Puts AI Extinction Risk Above 10%, Researcher Resigns Same Day

On September 9, two pieces of news about AI existential risk sparked intense discussion within and around Anthropic: Evan Hubinger, head of the alignment science team, publicly quantified his own estimate of extinction risk, while a pretraining researcher announced his resignation and slammed the frontier labs' race pace. Together, they once again exposed the deep tension inside frontier AI labs of "building while warning."

Confirmed

• Evan Hubinger (head of alignment science at Anthropic) publicly stated that the people building AI "sincerely believe" AI could kill all of humanity before the end of this decade—this is not marketing talk; he personally estimates the probability within the next decade at more than 10%.
• Hubinger acknowledged Anthropic is doing its best to reduce risk, but admitted there is currently no solution to the superintelligence alignment problem and no clear path to one, warning that the company is not on track.
• Researcher Jacob Coxon (@hilbertspaess) announced his resignation from Anthropic the same day. Over the past three years he worked on pretraining research at both OpenAI and Anthropic; in his statement he criticized both companies for not acting responsibly and "racing straight toward self-improving superintelligence, betting our lives on it."
• The story was covered by Forbes, CNBC, WSJ, and other outlets, and related Reddit threads sparked wide-ranging debate about genuine beliefs in AI risk versus doom-mongering.

Unconfirmed

• A Polymarket account claimed Coxon warned superintelligence "could kill us all by the end of this decade"—a secondhand news summary with no direct corroboration from his own long-form statement.

Why it matters

This is a rare quantified statement about extinction-level risk from a core figure at a frontier lab, and it dovetails with the same-day public departure of a safety-minded researcher—showing genuine internal disagreement at Anthropic over the pace of the AGI race and safety commitments, and reigniting outside debate over whether labs "say one thing and do another."

2026-09-09 ~ 2026-09-09 · 70 related posts

• WSJ: Anthropic researcher quits AI industry over fears of uncontrollable AGI race — peterwildeford · 2026-09-09
• Researcher quits Anthropic after 3 years in pretraining at OpenAI and Anthropic, blasting both labs — peterwildeford · 2026-09-09
• Exec quits over industrywide rush to build self-improving AI, citing humanity-ending risk — Miles_Brundage · 2026-09-09
• Anthropic researcher Jacob Coxon quits AI over fears self-improving models could go uncontrollable by 2027 — rohanpaul_ai · 2026-09-09
• Anthropic researcher Jacob Coxon quits AI over fears self-improving models could go uncontrollable by 2027 — rohanpaul_ai · 2026-09-09
• Anthropic Researcher Quits Over AI Fears, WSJ Reports — Bubbly-Air7302 · 2026-09-09
• Anthropic researcher Jacob Coxon resigns, warning superintelligence could kill us all by 2030 — Polymarket · 2026-09-09
• [source] Anthropic researcher: >10% chance AI kills all humans within a decade, alignment unsolved — EvanHub · 2026-09-09
• Anthropic Researcher Jacob Resigns, Sparking Sarcastic Debate Over AI Safety Exit Strategy — ctjlewis · 2026-09-09
• Anthropic staff put AI extinction odds at 'Jets make the playoffs' level — gottapatchemall · 2026-09-09
• "The People Building AI Seriously Believe It Could Kill Us All by 2030" — Puzzled-Ad-6854 · 2026-09-09
• Self-described Anthropic employee resigns, claims AI will 'probably kill us all' — AIandDesign · 2026-09-09
• Anthropic researcher quits over superintelligence race as 1,386 lab employees sign statement urging US to help pace frontier AI — EvanHub · 2026-09-09

37 near-duplicate retellings: Miles_Brundage · peterwildeford · AGI Hunt · AICopyLab · Miles_Brundage · EvanHub · ramagetime · ccerrato147 · JosephJacks_ · AccBalanced · Polymarket · kevinnbass · AndyMasley · EvanHub · JeffLadish · sjgadler · JeffLadish · sahilypatel · sjgadler · SydSteyerhart · ctjlewis · birchlse · sjgadler · iamfakhrealam · builderjaydub · SIGKITTEN · builderjaydub · Delahuntagram · austinc3301 · austinc3301 · AGI Hunt · Tough_Control2052 · tomchapin · WonderFactory · rkulidzan · 新智元 · JFPuget