Anthropic safety researchers resign over catastrophic risk concerns

Two key safety researchers at Anthropic, including alignment lead Evan Hubinger and team member Jacob Coxon, have stepped down from their roles. Their departure comes alongside public warnings regarding the potential dangers of rapidly advancing artificial intelligence. Hubinger cautioned that building increasingly capable models without strict oversight creates a plausible threat to human survival within the next ten years.

Anthropic was originally created with a explicit focus on building safe, controllable systems, which makes high profile departures over safety concerns particularly notable. When researchers whose primary job is preventing dangerous model behavior choose to walk away, it indicates a deepening friction between the rush to launch commercial tools and the technical work required to keep them secure. For everyday people, these resignations suggest that even labs dedicated to caution may be under pressure to prioritize speed over safety.

It is still unclear whether these specific departures stem from disagreement over an upcoming product release or a broader disillusionment with industry standards. As market competition pushes companies to accelerate their launch schedules, a crucial open question is whether self-regulation by AI developers can ever be sufficient. Without greater independent oversight, the public has little way of knowing if existing safety protocols are keeping pace with the systems being built behind closed doors.