DeepMind safety researcher leaves to join independent evaluator METR, declining OpenAI and Anthropic offers

Engels, a former member of Google DeepMind's AGI safety team, announced on September 13 that he had left the company three weeks earlier and turned down offers from OpenAI and Anthropic to join the independent evaluation organization METR. He cited excessive technological risk from recursive self-improvement and AI loss of control, arguing that frontier AI companies are racing toward superintelligence while humans still cannot prove a model is safe enough to enter such a self-improvement loop.