Sören Mindermann

Sören Mindermann is a machine learning researcher and AI safety scientist currently based in Montreal, where he is a postdoctoral researcher at Mila (Quebec Artificial Intelligence Institute) under Yoshua Bengio. He completed his DPhil (PhD) in machine learning at the University of Oxford (2019-2023), supervised by Yarin Gal at the OATML group and Allan Dafoe at the Centre for the Governance of AI, co-funded by Oxford and Google DeepMind. He also holds degrees in machine learning from UCL and in mathematics and Future Planet Studies from the University of Amsterdam. He served as the Scientific Lead of the first International AI Safety Report (2025), a comprehensive review of AI capabilities and risks backed by 33 nations, and is a Research Affiliate at the Oxford Martin AI Governance Initiative. His research covers AI safety evaluations, large language model honesty, data selection for large-scale deep learning, causal inference, and health applications of machine learning. Notable publications include co-authorship on "The Alignment Problem from a Deep Learning Perspective" (ICLR 2024), "Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training" (2024), and "Managing Extreme AI Risks amid Rapid Progress" (Science, 2024), as well as influential COVID-19 policy intervention studies published in Science and Nature Communications. He received funding from the Long-Term Future Fund for his AI strategy PhD at Oxford/FHI.

No endorsements yet.

Personal Website

https://www.soren-mindermann.com/

LinkedIn

LinkedIn

Twitter / X

Twitter

LessWrong

-

EA Forum

-