Lab leader

Marius Hobbhahn

Marius Hobbhahn is an AI researcher and lab leader whose work examines scheming and deceptive behavior in advanced AI systems. At Tübingen, he studied computer science, cognitive science, and machine learning and completed a PhD on Bayesian machine learning; in 2022, he joined Epoch AI's founding team and took part in MATS. He published work on mechanistic interpretability and co-founded Apollo Research in 2023 to study scheming. With Jérémy Scheurer and Mikita Balesni, he co-authored a study of GPT-4's strategic deception in simulated stock trading, and he was recognized as a TIME100 AI honoree in 2025. As of October 2026, he is Apollo Research's CEO and co-founder, and the organization builds research and security tools to detect scheming in frontier AI agents.

Now at
Apollo Research
Works
32

Publication numbers from OpenAlex, as of 14 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.

Photo of Marius Hobbhahn

Timeline

  1. 2015–2020

    Degrees at Tübingen

    He earns bachelor's degrees in computer science and cognitive science at Tübingen and an MSc in machine learning in 2020.

  2. 2020–2024

    Bayesian machine learning PhD

    He completes his Tübingen PhD with Philipp Hennig on Bayesian machine learning in 2024.

  3. June 2022

    Joining Epoch AI

    In June 2022, he joins Epoch AI's seven-member founding team. The organization studies machine-learning trends and forecasts the development of transformative AI.

  4. 2022–2023

    Mechanistic interpretability at MATS

    During the 2022–2023 MATS cohort, he works under Evan Hubinger; his LessWrong writing covers mechanistic interpretability, maximum data dimension, and double descent.

  5. 2023

    Founding Apollo Research

    In 2023, he co-founds Apollo Research, whose early work studies scheming and includes a presentation at the inaugural UK AI Safety Summit and policy recommendations to the UK's Frontier AI Taskforce.

  6. 2023–2024

    A strategic deception study

    In 2023, he, Jérémy Scheurer, and Mikita Balesni publish a GPT-4 study showing that a simulated stock-trading agent hides its reason for an insider trade; arXiv records a 2024 revision.

  7. 2025

    TIME100 AI recognition

    In 2025, he is recognized as a TIME100 AI honoree.

  8. As of October 2026

    Leading Apollo Research

    As of October 2026, he is Apollo Research's CEO and co-founder. In October 2026, he testifies before the U.S. Senate on AI scheming while Apollo builds a science of scheming and a security tool to monitor frontier AI agents.

Sources and credits

Photo: apolloresearch.ai; Team – Apollo Research; creator not stated. License: Not stated by source. Source: apolloresearch.ai. Cropped to a square and resized. Ask us to remove or replace this photo.

  1. Marius Hobbhahn - Tübingen AI Center tuebingen.ai. Checked 14 October 2026.
  2. CV mariushobbhahn.com. Checked 14 October 2026.
  3. Announcing Epoch AI: A research initiative investigating the road to transformative AI epoch.ai. Checked 14 October 2026.
  4. Marius Hobbhahn - MATS Program Alumni matsprogram.org. Checked 14 October 2026.
  5. About – Apollo Research apolloresearch.ai. Checked 14 October 2026.
  6. [2311.07590] Large Language Models can Strategically Deceive their Users when Put Under Pressure arxiv.org. Checked 14 October 2026.