An Anthropic safety researcher says the odds of AI wiping out humanity are north of 10 percent - and that estimate landed the same day a colleague quit over the company's approach to safety.
The unnamed researcher's figure surfaced just hours after Jacob Coxon announced his resignation on X. Coxon, who has trained AI systems at both Anthropic and OpenAI, accused Anthropic and its rivals of racing straight toward self-improving superintelligence and gambling with people's lives. He called the industry's rush to build systems smarter than humans careless. The researcher's own estimate puts the chance that AI kills all humans at more than 1 in 10 by the end of the decade.
This isn't an outside critic throwing out a scary number. It's someone inside one of the field's most safety-focused labs, speaking up the same day a trainer walked out over the exact same worry. When people building the technology assign it double-digit extinction odds, the industry's usual line about proceeding carefully gets harder to take at face value.
It's also not an isolated incident - safety-driven exits have dogged OpenAI too, and Anthropic has built its entire brand on being the cautious alternative. A double-digit doomsday estimate from inside that house is not a great look.