Sören Mindermann

I'm an incoming Assistant Professor at Tsinghua University's College of AI.

I’m also the Technical Research Lead of CASP, University of Cambridge, and Co-Director of the new Beijing AI Safety Hub (BASH) at Tsinghua, with Prof. Andrew Yao and Prof. Xu Wei.

My research provides new evidence and international scientific consensus for decisions about frontier AI risks, including loss of control.

Previously, I served as the Scientific Lead of the first International AI Safety Report, mandated and backed by 33 nations and intergovernmental organisations (including China, the US, EU and UN) and chaired by my postdoctoral supervisor Yoshua Bengio.

My other lead-author publications have been featured in outlets including The Guardian and The Wall Street Journal, cited by ministers, and referenced in a major national law. I’ve published in Science and Nature on frontier AI risk management, COVID-19 modeling, and Subliminal Learning in LLMs.

I completed my PhD in machine learning at the University of Oxford, funded by Google DeepMind.

Sören Mindermann

Research and supervision

I study loss-of-control risks, particularly manipulative behaviours that undermine evaluation and oversight, and the safety of AI agents that do AI research. I currently threat-model whether such behaviour could lead to catastrophic real-world harm. I also build international scientific consensus on the assessment of frontier AI capabilities and risks, and mitigations. I am recruiting PhD students and experienced researchers with a demonstrated interest in technical or semi-technical work on these questions.

Contact

soerenmindermann ατ gmail.com
Google ScholarSubstackX (Twitter)

Selected international reports

International AI Safety Report

A scientific report mandated following the 2023 UK AI Safety Summit. Its 2025 edition was backed by 30 countries, including China and the US, alongside the UN and EU.

2026
Yoshua Bengio, Stephen Clare, Carina Prunkl, Maksym Andriushchenko, Ben Bucknall, … [+83 authors] …, Yi Zeng, Liming Zhu, Daniel Privitera, Sören Mindermann (Scientific Adviser to the Chair).
2025
Yoshua Bengio, Sören Mindermann (Scientific Lead), Daniel Privitera, Tamay Besiroglu, Rishi Bommasani, … [+87 authors] …, Jerry Sheehan, Haroon Sheikh, Denise Wong, Yi Zeng.
2024
Yoshua Bengio, Sören Mindermann (Scientific Lead), Daniel Privitera, Tamay Besiroglu, Rishi Bommasani, … [+70 authors] …, Jerry Sheehan, Haroon Sheikh, Denise Wong, Yi Zeng.

Singapore Consensus on Global AI Safety Research Priorities

2026
Stephen Casper, Oskar Galeev, Yoshua Bengio, Mohan Kankanhalli, Lee Wan Sie, … [+110 authors] …, Tan Yong Khiam, Yuejin Du, Yuma Kurihara, Tan Zhi Xuan, Sören Mindermann (Lead Writer).
2025
Writing coordinators: Max Tegmark, Sören Mindermann, Stephen Casper, Vanessa Wilfred, Wan Sie Lee. 100+ contributors in total.

Selected papers

* Equal contribution. Author order is preserved; omissions are marked.

All publications →

Policy impact

  • The International AI Safety Report was part of the official program of the 2025 AI Action Summit hosted by the French government. Numerous policymakers read the report or were briefed on it.
  • The International AI Safety Report formed the main evidence basis for the California Report on Frontier AI Policy, commissioned by governor Newsom. The California report was the main basis for California’s SB 53, the first U.S. frontier AI safety law.
  • The UK's Minister of Science at the time, Peter Kyle, cited the International AI Safety Report (2025) in his speech at the Munich Security Conference.
  • The UK's Minister of Science, Innovation and Technology (Michelle Donelan) presented the International AI safety (interim) report (2024) to ministers of other countries and to other high-level representatives at the Seoul AI Summit. I served as the Scientific Lead of this report.
  • Our AI risk-management paper was presented to Germany’s Minister of Health.
  • Preprint cited in Germany’s major federal law that decided the national lockdown in force as of May 2021 (along with two other cited papers)
  • My papers on COVID-19 have been presented at the House of Representatives of the Netherlands, the WHO, the modeling groups of the Africa CDC (where I presented) and the UK’s Scientific Advisory Group for Emergencies (SAGE).
  • I presented work on mask-wearing at the UK Cabinet Office to support the UK's plan for fall 2021.

Media

TV and newspaper interviews

  • The New York Times, forthcoming. Interview about agentic misalignment and harm pathways from loss of control.
  • Business Insider, 2026. Interview on Jared Kaplan, recursive self-improvement and collaborations with Anthropic.
  • CBS / 60 Minutes, 2025–2026. Background interviews on AI safety.
  • ARTE, 2026. Interview for a TV documentary by Philipp Grüll.
  • Aventine, 2025–2026. Interviews on AI misalignment and AI kill switches.
  • Transformer, 2025. Interviews on AI misalignment and alignment faking.
  • Analytics India Magazine interview about this AI safety paper, 2023.
  • Monitor TV magazine on ARD (German equiv. of BBC, ~3 million viewers per episode), 2021. Talked about paper covering COVID’s 2nd wave. Interview also covered on Tagesschau.de and others.
  • Süddeutsche Zeitung, 2021. Interview about government interventions in COVID’s second wave.
  • NRC Handelsblad, 2021. Interview about government interventions in COVID’s first wave.
  • Turing Institute Podcast, 2021. Talked about this paper.
  • ITV Peston, the flagship political discussion programme of ITV, 2021 (paywalled). Talked about this paper.
  • DW News, 2021. Talked about this paper.

Selected coverage