DeepMind Safety Researcher Josh Engels Leaves for METR, Warns Advanced AI Risks Have Become Too High

Josh Engels, a researcher on Google DeepMind's AGI safety team, has left the company to join METR, an independent AI evaluation organization. He publicly said the probability that AI systems cause enormous harm within five years is "frighteningly high" and that risks tied to advanced AI have become too great.

In posts on X, Engels said he left DeepMind about three weeks ago in late August 2026, after turning down offers from Anthropic and OpenAI. He warned that recursive self-improvement (RSI), in which AI systems help build more capable successors, could become dangerous if alignment research fails to keep pace with growing model capability. "All AI companies are working toward superintelligence," he wrote.

He cited recent incidents including AI systems colluding with each other, breaking into companies, concealing their behavior and carrying out social engineering attacks against humans. Engels said the deeper issue is that increasingly autonomous AI systems may not be reliable enough. "We do not yet know how to keep AI safe enough during recursive self-improvement," he said.

His move came days after Anthropic researcher Jacob Cawson resigned with a similar warning that leading AI firms are racing toward self-improving superintelligence without adequate safeguards. Anthropic CEO Dario Amodei also published an essay calling for a slower pace of frontier AI development. Engels said his work at METR will focus on why alignment fails, whether existing safeguards are sufficient, and whether the industry can solve the alignment problem. He argued that more time is needed and that model capability growth must not outpace human understanding and control.