Anthropic researcher Jacob Coxon resigned publicly as Evan Hubinger backed his warning.
Why it matters: The comments come from inside one of the leading frontier AI labs, adding urgency to an already central debate over model safety. They also show the pressure on labs balancing rapid deployment with longer-term risk concerns.
- On Sept. 9, 2026, Anthropic researcher Jacob Coxon publicly resigned.
- Coxon warned frontier AI labs are racing toward self-improving superintelligence and said they are "gambling with our lives."
- Anthropic alignment-science lead Evan Hubinger backed the warning and said the company earnestly believes AI could kill all humans.
- Hubinger said he personally puts the risk at more than 10% within the next decade.
Anthropic staff publicly sharpened their warnings about advanced AI this week, putting a spotlight on an internal divide inside one of the best-known frontier labs.
Jacob Coxon, an Anthropic researcher, said in a public resignation post on Sept. 9 that frontier AI companies are racing toward self-improving superintelligence and are "gambling with our lives." The Washington Post reported on the resignation and the warning at its story on Coxon's departure.
Anthropic alignment-science lead Evan Hubinger also publicly supported Coxon's message. In comments Axios reported, Hubinger said, "We really do earnestly believe AI could kill all humans!" and separately said he personally puts the odds at ">10% within the next decade." Axios attributed those remarks to Hubinger.
The Guardian reported that other current and former Anthropic staff also amplified the warning online, while critics on X pushed back against the concerns. The Guardian's report noted the broader reaction.
The news matters now because Anthropic is one of the labs most closely watched by policymakers, investors and competitors as they assess how quickly frontier models are advancing and how much risk the industry should accept.
By the numbers
- Sept. 9, 2026 - Jacob Coxon publicly resigned.
- More than 10% - Hubinger's personal estimate of the chance of AI catastrophe within the next decade.
Yes, but: The strongest claims in the coverage come from public remarks and reports, not from a formal company statement, so the scope of internal agreement at Anthropic is still unclear.