Josh Engels announced on X that he left Google DeepMind's AGI Safety team to join METR as a Member of Technical Staff. He said the move took place about three weeks before his September 12 announcement.
At METR, Engels is investigating where misaligned behavior comes from during training, whether current safeguards are sufficient, and whether the field is on track to solve alignment. He argued that independent organizations are needed to evaluate frontier systems and hold developers accountable.
Engels previously researched how to understand the internal workings of language models, including whether sparse autoencoders can reveal meaningful model features. His recent work also examines scalable oversight—how people might supervise systems that perform tasks beyond human ability.