Founding Apollo Research member; co-discovered the Reversal Curse.
Why chosen
He connects scheming evaluations with the race to keep AI reasoning readable. As a MATS scholar with Owain Evans he co-discovered the Reversal Curse, showing that models trained on "A is B" fail to learn "B is A." In 2023 he became a founding member of Apollo Research, building evaluations for scheming and situational awareness, and in 2025 was co-first author, with Tomek Korbak, of the cross-lab paper on chain-of-thought monitorability. He then joined OpenAI to work on alignment, until October 2026. He noticed the models noticing.
LeftDeparted OpenAI in October 2026; co-wrote an open letter calling for clear rules on outside safety collaboration. Authorship confirmed on arXiv (2507.11473); OpenAI role confirmed by TechCrunch (Oct 8, 2026).
