Anthropic Researcher Exit Reignites AI Extinction Debate

Why this matters

Evan Hubinger runs Alignment Science at Anthropic. When he puts human extinction odds above 10%, that's not a Twitter troll or a doomer blogger, that's someone whose job is literally to assess this risk at one of the labs building frontier models. Jacob Coxon's exit post and Hubinger's reply matter because they collapse the distance between "safety talk as marketing" and "safety talk as genuine internal belief." For developers and founders building on top of Claude or competing models, this is worth sitting with: the people closest to the technical guts of these systems aren't uniformly reassuring, even inside the company that has staked its brand on being the careful one.

We'd push back on treating any single percentage as gospel, extinction-risk estimates are notoriously squishy and contested even among safety researchers. But the fact that this argument is happening in public, sparked by an internal resignation rather than an outside critic, tells you something about where confidence levels actually sit inside frontier labs right now. Watch whether other Anthropic staff back Hubinger's number or distance themselves from it.