PeopleofSuperintelligence

Researcher

Thilo Hagendorff

Thilo Hagendorff is an AI researcher who studies AI ethics and the behavior of language models. He pursued philosophy, cultural studies and German literature at Konstanz and Tübingen and earned a sociology doctorate at Tübingen; a philosophy essay of his appeared in Juventas in 2011. His research moved from comparing AI ethics guidelines in 2020 and proposing machine psychology in 2023 to mapping generative-AI ethics and testing language-model deception in 2024, then benchmarking latent-space reasoning and studying Stable Diffusion bias in 2025. As of October 2026, he leads an AI Safety group at the University of Stuttgart; earlier, he worked in Tübingen’s Machine Learning cluster and held visiting positions at Stanford, UC San Diego and ELLIS Alicante.

Timeline

  1. 2011

    Studies and a philosophy essay

    He studies philosophy, cultural studies and German literature at the universities of Konstanz and Tübingen, and earns a sociology doctorate at Tübingen. In 2011, he publishes “Sprachlicher Idealismus und das Problem der Sprache-Welt-Referenzrelationen” in Juventas.

  2. 2020

    An AI ethics survey

    In 2020, he publishes an overview in Minds & Machines that compares AI and machine-learning ethics guidelines, identifies shared principles and gaps, and considers how the guidelines can work in practice.

  3. 2023

    Machine psychology

    In 2023, he proposes machine psychology: psychological behavioral experiments to study language models’ reasoning and behavior.

  4. 2024

    Mapping generative-AI ethics

    In 2024, he publishes a scoping review of generative-AI ethics in Minds and Machines, mapping 378 normative issues across 19 topic areas.

  5. 2024

    Language-model deception

    In 2024, his PNAS paper reports experiments showing that state-of-the-art language models understand deception strategies and can induce false beliefs in other agents.

  6. 2025

    Latent-space reasoning

    In 2025, with Sarah Fabi, he introduces a 4,000-item benchmark for measuring model-internal reasoning across different domains in “Beyond Chains of Thought.”

  7. 2025

    Stable Diffusion bias

    In 2025, with Matthias Schneider, he tests ten popular Stable Diffusion models and finds harmful prompt responses, including biased portrayals of Black people in violent contexts.

  8. As of October 2026

    AI Safety in Stuttgart

    As of October 2026, he leads an independent AI Safety research group within the University of Stuttgart’s Stuttgart Research Focus Interchange Forum for Reflecting on Intelligent Systems. His research examines AI-agent behavior, with work on deception, emergent misalignment, evaluation awareness and shutdown resistance. Earlier, he works as a postdoctoral researcher in Tübingen’s Cluster of Excellence Machine Learning: New Perspectives for Science and holds visiting positions at Stanford, UC San Diego and ELLIS Alicante.

Career

  1. NowUniversity of StuttgartAI Safety group leader
  2. University of TübingenPostdoctoral researcher
  3. Stanford UniversityVisiting scholar
  4. University of California, San DiegoVisiting scholar
  5. European Laboratory for Learning and Intelligent SystemsVisiting scholar

Education

  1. University of Konstanz
  2. University of TübingenSociology doctorate

From the biography.

  1. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionIncludes University of Stuttgart. We estimate where the people shaping AI studied and where their careers led. Massachusetts Institute of Technology leads with 467 former students, including 291 recorded doctorates.
  2. The company share of AI researchers’ top papers was 31% in 2020–2024 and 8% in 1970–1974Includes University of Stuttgart. We estimate where the people shaping AI worked when their most-cited papers appeared: the company share was 31% in 2020–2024 and 8% in 1970–1974, and Google led all organizations in 2020–2024.
  3. Google has the highest combined h-index among AI organizationsIncludes University of Stuttgart. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.
  4. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes Stanford University. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.

Sources and credits

Photo: We have no published photo of Thilo Hagendorff yet. Send us a free-licensed photo.

  1. Thilo Hagendorff – Wikipedia de.wikipedia.org. Checked 21 October 2026.
  2. Publications – Thilo Hagendorff thilo-hagendorff.info. Checked 21 October 2026.
  3. KI-Ethik - Überblickspapier analysiert Ethikrichtlinien und -prinzipien | Universität Tübingen uni-tuebingen.de. Checked 21 October 2026.
  4. [2303.13988] Machine Psychology arxiv.org. Checked 21 October 2026.
  5. [2402.08323] Mapping the Ethics of Generative AI: A Comprehensive Scoping Review arxiv.org. Checked 21 October 2026.
  6. Herr Dr. Thilo Hagendorff | Interchange Forum for Reflecting on Intelligent Systems (SRF IRIS) | Universität Stuttgart iris.uni-stuttgart.de. Checked 21 October 2026.
  7. [2307.16513] Deception Abilities Emerged in Large Language Models arxiv.org. Checked 21 October 2026.
  8. [2504.10615] Beyond Chains of Thought: Benchmarking Latent-Space Reasoning Abilities in Large Language Models arxiv.org. Checked 21 October 2026.
  9. Investigating toxicity and Bias in stable diffusion text-to-image models | Scientific Reports nature.com. Checked 21 October 2026.
  10. About me – Thilo Hagendorff thilo-hagendorff.info. Checked 21 October 2026.