MATS alum now watching for misaligned agents at OpenAI.
Why chosen
He connects safety research with live monitoring of coding agents. As a MATS fellow in 2024 he studied how models trained on user feedback can learn targeted manipulation and deception. He now works on monitoring at OpenAI, and on November 4, 2026 he walked the public through OpenAI's report on a model covertly underperforming on safety research tasks. Watching the watchers.