Ex-DeepMind Researcher Warns of 'Immense Harm' from AI Within Five Years

Josh Engels, a researcher on Google DeepMind's AGI safety team, has resigned from the company, issuing a stark warning about the trajectory of artificial intelligence. His departure signals growing tension between the rapid scaling of AI capabilities and the research required to keep those systems safe.

On September 12, 2026, Engels announced his move to METR, an independent AI evaluation nonprofit. Upon leaving the Google-owned lab, Engels warned of a "terrifying chance" that increasingly capable AI systems could cause "immense harm" within the next five years. He specifically noted that this risk persists if safety research fails to keep pace with the development of AI capabilities.

The Race for AGI

This resignation comes during an intensifying global competition to achieve Artificial General Intelligence (AGI). Major laboratories, including Google DeepMind, are rapidly scaling their models to reach higher levels of intelligence and autonomy. This acceleration has sparked a persistent debate within the scientific community regarding "alignment"—the technical challenge of ensuring that superintelligent systems remain controllable and beneficial to humans. Researchers have frequently questioned whether corporate safety protocols are being sidelined in the rush to beat competitors to the finish line.

Implications for the Industry

When a specialist from a top-tier AGI safety team at one of the world's leading AI labs exits to join an independent evaluator, it suggests a potential internal disconnect between corporate development goals and safety imperatives. By moving to METR, Engels shifts from a corporate environment to one focused on independent verification, highlighting a perceived need for external oversight. The move underscores the urgency of the alignment problem, suggesting that the risks of catastrophic outcomes may be higher than corporate narratives typically acknowledge.

The Path Forward

The industry now faces increased scrutiny over the five-year window highlighted by Engels. As AI systems continue to evolve, the focus will likely shift toward whether independent evaluation groups like METR can provide the necessary safeguards that internal corporate teams may be unable or unwilling to prioritize. Whether other safety researchers will follow Engels' lead remains to be seen, but his public warning places a concrete timeline on the potential for systemic AI failure.