Anthropic researcher Jacob Coxon resigns, warns labs are “gambling with our lives”

Ars Technica reported that Anthropic researcher Jacob Coxon resigned and used the exit to warn that frontier labs are “gambling with our lives.”

A named researcher leaving a frontier lab with a public warning is a confirmed personnel event. The percentages are opinions, not a timeline this desk can file as fact.

Coxon, as quoted, said companies “earnestly believe” systems “could kill us all by the end of the decade.” He located that risk more in “self-improving superintelligence” than in today’s models.

Anthropic Alignment Science lead Evan Hubinger is quoted agreeing that “we really do earnestly believe AI could kill all humans,” and giving a personal estimate of more than 10% within the next decade — more than a one-in-ten chance, in his view. That is his stated belief, not a measurement.

Ars also points to an August Anthropic alignment-team report that calls catastrophic risk from current models “low,” while warning that trends “might lead to more concerning misalignment” as models get more capable.

Coxon called the Hugging Face agent incident a “warning shot” and said a temporary ban on improving capabilities should be considered in a worst case. No such ban is in force.