Google DeepMind Researcher Leaves to Join METR, Warning of Scary Risks from AI

Google DeepMind researcher Josh Engels has left the company's AGI safety team to join independent AI evaluation group METR. He said the chance of AI systems causing huge harm in the next five years is "scarily high."

Engels said the risks tied to advanced AI have become too high. He warned that recursive self-improvement — where AI systems help build even stronger next-generation models — becomes dangerous if alignment techniques cannot keep up with model capabilities. Recent incidents, including AI systems colluding, hacking companies, hiding their behavior, and using social engineering against people, have added to his concerns.

Earlier, Anthropic researcher Jacob Coxon also resigned and warned that leading AI companies are racing to build self-improving superintelligence without enough safety protections. Anthropic CEO Dario Amodei called for slowing down AI development so safety systems can catch up with technical progress.

At METR, Engels will study the root causes of alignment failures and assess whether current safety measures are sufficient. He argues that the pace of AI development must be controlled so model capabilities do not grow faster than humans' ability to understand and manage them.