Can AI Kill All Humans Within Next Decade? Anthropic Scientist Says 'Over 10% Chance'

An Anthropic scientist just said out loud what a lot of people in artificial intelligence only whisper. 

Anthropic researcher Jacob Coxon, in a social media post after his resignation, said, "The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt."

Reacting to Jacob's post on X, Evan Hubinger, Alignment Science Lead, Anthropic, said, "Jacob is correct here; we really do earnestly believe AI could kill all humans! I personally think it is >10 per cent within the next decade,"

Also Read | Layoffs Are Rising In Tech, But These Skills Are Helping People Unlock Better Pay

Notably, this is Hubinger's own estimate. Not official Anthropic estimate.

The Post That Started It All

In a post on X, Jacob Coxon, a 27-year-old pretraining researcher who just announced he was quitting the AI industry altogether, said he no longer believes any single lab can safely build self-improving AI, no matter how careful it tries to be. 

Coxon has worked at both OpenAI and Anthropic. He accused the two companies of "racing straight to self-improving superintelligence and gambling with our lives."

Reacting to his post, Hubinger said, "I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to."

In a follow-up post, Hubinger tried to draw a line. He said today's Claude models pose low risk. The real fear, he wrote, is superintelligence emerging from recursive self-improvement -- AI that gets smarter by rewriting itself -- which he says is arriving faster than expected.

Internet Says 'Concerning'

Reactions to Coxon's "AI can kill us all" posts are already rolling in. 

Arun Rao, adjunct professor at UCLA and a builder of large-scale ML systems, sees this as a symptom of a wider ideological split inside AI labs. He wrote: "Whistleblower from Anthropic. Personally, I think his EA-influenced, doomer-keyed arguments are common views at Anthropic (at least a large minority), less common but moderated at OpenAI (more about pacing to insure safety), and not that common at other labs (eg Deepmind, Meta, GLM, DeepSeek, etc)."

Meanwhile, US Congress Member Ted Lieu wasted no time turning it into a policy argument: "Below is exhibit number 739 for why we need to pass the bipartisan AI Kill Switch Bill ASAP."

The number is obviously rhetorical -- but the point isn't. Lieu has been pushing for hard legal guardrails on frontier AI, and every high-profile warning from inside the labs becomes fuel for that push in Washington.

Bill Ackman, CEO of Pershing Square, kept it to one word: "Concerning."