Dossier · Ongoing
Safety researchers leave the AI labs
From Jacob Coxon's resignation at Anthropic to two safety researchers joining METR in September 2026: the departures from the AI labs, tracked over time.
In September 2026, several safety researchers leave the leading AI labs within days of each other – and publicly cite the pace of development and the lack of independent oversight as their reasons. Pretraining researcher Jacob Coxon opens the series: he resigns from Anthropic on September 8 and accuses OpenAI and Anthropic alike of a reckless race toward self-improving superintelligence; Anthropic’s alignment lead Evan Hubinger then confirms a double-digit risk of a fatal AI failure within ten years.
A few days later, Joe Benton, previously head of the Scalable Oversight team at Anthropic, and Josh Engels of Google DeepMind move to the independent auditing organization METR. Both criticize that transparency about AI risks so far remains purely voluntary. This dossier collects the departures and the companies’ responses – and tracks whether the warnings lead to binding oversight mechanisms.