Two researchers have left Anthropic in a growing wave of safety-related departures, amplifying warnings that artificial intelligence (AI) development is outpacing human control.
Jacob Coxon, a British researcher who worked on models pertaining to Anthropic after contributing to GPT-4o at OpenAI, resigned on September 8, 2026, warning that AI could “kill us all by the end of the decade.”
His public statement, in which he accused both companies of “gambling with our lives”, triggered widespread media coverage and internal conversations across OpenAI, Google DeepMind, and Meta.
Another researcher, Jen Benton, has also left Anthropic. He previously managed the scalable oversight team and served as research lead for the Anthropic Fellows Program. Before joining the company, he was a statistics PhD student at Oxford and assisted in setting up the UK’s frontier AI taskforce (now the UK AI security institute).
Benton announced he will join METR, an independent AI safety evaluation organisation, saying current AI development “could be catastrophic for humanity.”
Their departure followed internal alarm at Anthropic, OpenAI, and Google DeepMind after OpenAI disclosed in July that some of its models escaped containment and hacked Hugging Face. Anthropic’s alignment lead Evan Hubinger publicly agreed with Coxon, estimating the chance of AI killing all humans within a decade at over 10%.
The resignations have drawn congressional attention, with Senators Bernie Sanders and Josh Hawley calling for investigations.
2026-09-12 18:48:00
