Josh Engels joins METR and warns alignment work is falling behind
The former Google DeepMind safety researcher warns of immense AI harm within five years and calls for pacing and independent investigation.
Josh Engels says he left Google DeepMind's AGI safety team about three weeks before his September 12 announcement and is joining METR. His personal website describes previous work on DeepMind's AGI safety team and MIT interpretability research, and his current focus on incident investigations at METR. Josh Engels Josh Engels
Engels argues that alignment is not keeping pace with capabilities and is particularly concerned about recursive self-improvement. He fears immense harm from AI within five years, without giving a numerical probability. This is his personal assessment, not a METR forecast or a measured comparison of alignment across models. Josh Engels
He calls for slower development and more independent investigation. His proposed work concerns misalignment arising during training and whether mitigations can adequately address it. He also encourages additional independent organizations to work on the problem. Josh Engels
Sources & attribution
- Firsthand account 12 Sept 2026METR move and alignment warning
Josh Engels. Original personal announcement and warning; complete post inspected.
- Firsthand account Accessed 12 Sept 2026Josh Engels — About
Josh Engels. Author's biography identifies current METR work and prior DeepMind and MIT research. A self-description, not an employer personnel record.