
Josh Engels resigned from Google's AGI safety team to join METR, warning that advanced AI systems could cause severe harm if development outpaces safety alignment.
Engels expressed concern over recursive self-improvement, where AI systems build increasingly capable successor systems without verified safety guarantees.
At METR, Engels will study alignment failures during training and evaluate whether current mitigations are sufficient for managing autonomous capabilities.