Why an Anthropic Safety Lead Says AI Could Kill All Humans Within a Decade
Anthropic's Alignment Science Lead Evan Hubinger has publicly put his personal estimate of AI causing human extinction within ten years at greater than 10%, backing a resigning colleague who accused both Anthropic and OpenAI of racing toward self-improving superintelligence without acting responsibly.
The person whose job is stress-testing Anthropic's safety work just put a number on his own fear. Evan Hubinger, the company's Alignment Science Lead, wrote on X that he and colleagues "really do earnestly believe AI could kill all humans" — and pegged his personal estimate at greater than 10% within the next decade.
He added that Anthropic is trying its best, but has no plan yet to solve alignment for superintelligence and is not clearly on track to find one.
The comment came in response to Jacob Coxon, a pretraining researcher who resigned from Anthropic this week after previously working at OpenAI. The details that give the exchange weight:
- Coxon accused both companies of "racing straight to self-improving superintelligence" and insisted the warnings were not a marketing stunt
- Hubinger later clarified that risk from current models is low — his concern is recursive self-improvement, where a system designs its own successor without human input
- Roughly 1,400 researchers, including Anthropic co-founders Dario Amodei and Jared Kaplan and OpenAI's Jakub Pachocki, signed a July letter urging governments to help pace frontier AI development
The political response is already forming. Rep. Jay Obernolte and Rep. Lori Trahan introduced the FRONTIER Act to govern advanced model deployment, while Sen. Bernie Sanders and Rep.
Greg Casar have proposed a Ban Artificial Superintelligence Act, and in the UK former minister Darren Jones has called for a multinational treaty. It's worth reading Hubinger's figure for exactly what it is: a personal probability estimate, not a company forecast or a scientific consensus.
Plenty of researchers dispute both the timeline and the premise. But when the person hired to find the holes in a lab's safety methods says publicly that the holes aren't closed, that's a data point regulators are unlikely to ignore.
