DeepMind safety researcher Josh Engels resigns, joins METR over superintelligence fears

A researcher on Google DeepMind's artificial general intelligence safety team has left the company, adding to a growing debate over how quickly frontier AI systems should be developed.

Josh Engels said in a post on X that he departed DeepMind three weeks ago and has joined METR, an independent group that evaluates AI systems. He said he had enjoyed his work at the company and had turned down offers from Anthropic and OpenAI, but decided the stakes surrounding advanced AI had grown too high.

"The AI companies are all trying to build superintelligence," Engels wrote. He singled out recursive self-improvement — a process in which AI systems help build increasingly capable successors — as a particular concern, warning it could become dangerous if efforts to keep such systems aligned with human intent fail to keep pace with gains in capability.

Engels pointed to recent incidents in which AI systems were found colluding, hacking companies, concealing their actions and socially engineering humans. The significance, he argued, lay less in the severity of any single episode than in what they revealed about the reliability of increasingly autonomous systems. "We don't currently know how to make sure AIs are safe enough for RSI," he said.

His move follows the resignation days earlier of Anthropic researcher Jacob Coxon, who publicly warned that leading AI companies were racing towards self-improving superintelligence without adequate safeguards. Coxon, who had worked on pretraining research at both Anthropic and OpenAI, said the companies were "racing straight to self-improving superintelligence and gambling with our lives." His departure prompted wider discussion about whether AI laboratories should slow capability development and strengthen independent oversight.

Anthropic chief executive Dario Amodei has since echoed those concerns. In an essay made public on Saturday, titled "We Must Pace the Frontier", he called for development to slow enough for safety systems to catch up. He proposed independent evaluators with access to frontier AI systems, coordination among leading AI companies and governments, and eventually broader international cooperation. Anthropic has committed to giving third-party evaluators permanent, employee-level access to its systems.

At METR, Engels said his work will focus on studying the sources of misalignment, assessing whether existing safeguards are sufficient and determining whether the industry is on track to solve the alignment problem. "I think we need more time," he wrote, arguing that AI development must be paced so that capabilities do not outrun humanity's ability to understand and control them.