Thilo Hagendorff

A complete list of my publications as well as books can be found here. The following is a selection:

• Florin, Lucas; Knecht, Amelie; Schaller, Ulysse; Hagendorff, Thilo (2026): Deception by Omission: Language Models Knowingly Hide Their Mistakes. In arXiv:2610.11351, pp. 1–58. (Link)
• Vaugrante, Laurène; Hagendorff, Thilo (2026): How Much Do LLM-as-a-Judge Design Choices Matter? A Systematic Comparison of Prompt Designs, Rating Scales, and Models. In arXiv:2610.05094, pp. 1–34. (Link)
• Knecht, Amelie; Schaller, Ulysse; Summerfield, Christopher; Hagendorff, Thilo (2026): Shutdown Sabotage Propensities in Multi-Agent Systems. In arXiv:2609.28274, pp. 1–38. (Link)
• Knecht, Amelie; Florin, Lucas; Hagendorff, Thilo (2026): Evaluation Awareness in Language Models Has Limited Effect on Behaviour. In arXiv:2605.05835, pp. 1–29. (Link)
• Vaugrante, Laurène; Carlon, Francesca; Menke, Maluna; Hagendorff, Thilo (2026): Compromising Honesty and Harmlessness in Language Models via Deception Attacks. In Transactions on Machine Learning Research, pp. 1–29. (Link)
• Vaugrante, Laurène; Weckauff, Anietta; Hagendorff, Thilo (2026): Emergently Misaligned Language Models Show Behavioral Self-Awareness That Shifts With Subsequent Realignment. In arXiv:2602.14777, pp. 1–33. (Link)
• Hagendorff, Thilo; Derner, Erik; Oliver, Nuria (2026): Large reasoning models are autonomous jailbreak agents. In Nature Communications 17, pp. 1–8. (Link)
• Jotautaitė, Monika; Caviola, Lucius; Brewster, David A.; Hagendorff, Thilo (2025): Speciesism in AI: Evaluating Discrimination Against Animals in Large Language Models. In arXiv:2508.11534, pp. 1–29. (Link)
• Hagendorff, Thilo (2025): On the Inevitability of Left-Leaning Political Bias in Aligned Language Models. In arXiv:2507.15328, pp. 1–11. (Link)
• Hagendorff, Thilo; Fabi, Sarah (2025): Beyond Chains of Thought: Benchmarking Latent-Space Reasoning Abilities in Large Language Models. In arXiv:2504.10615, pp. 1–12. (Link)
• Vaugrante, Laurène; Carlon, Francesca; Menke, Maluna; Hagendorff, Thilo (2025): Compromising Honesty and Harmlessness in Language Models via Deception Attacks. In arXiv:2502.08301, pp. 1–14. (Link)
• Vaugrante, Laurène; Niepert, Mathias; Hagendorff, Thilo (2024): A Looming Replication Crisis in Evaluating Behavior in Language Models? Evidence and Solutions. In arXiv:2409.20303, pp. 1–23. (Link)
• Hagendorff, Thilo (2024): Mapping the Ethics of Generative AI. A Comprehensive Scoping Review. In Minds and Machines 34 (39), 1–27. (Link)
• Hagendorff, Thilo; Dasgupta, Ishita; Binz, Marcel; Chan, Stephanie C. Y.; Lampinen, Andrew; Wang, Jane X. et al. (2024): Machine Psychology. In arXiv:2303.13988, pp. 1–17. (Link)
• Hagendorff, Thilo (2024): Deception abilities emerged in large language models. In Proceedings of the National Academy of Sciences 121 (24), 1-8. (Link)
• Meding, Kristof; Hagendorff, Thilo (2024): Fairness Hacking: The Malicious Practice of Shrouding Unfairness in Algorithms. In Philosophy & Technology 37 (1), pp. 1–22. (Link)
• Hagendorff, Thilo; Fabi, Sarah; Kosinski, Michal (2023): Human-like intuitive behavior and reasoning biases emerged in large language models but disappeared in ChatGPT. In Nature Computational Science 3 (10), pp. 833–838. (Link)
• Hagendorff, Thilo; Fabi, Sarah (2023): Why we need biased AI: How including cognitive biases can enhance AI systems. In Journal of Experimental & Theoretical Artificial Intelligence, pp. 1–14. (Link)
• Hagendorff, Thilo; Bossert, Leonie N.; Tse, Yip Fai; Singer, Peter (2023): Speciesist bias in AI: how AI applications perpetuate discrimination and unfair outcomes against animals. In AI Ethics 3 (3), pp. 717–734. (Link)
• Hagendorff, Thilo (2022): A Virtue-Based Framework to Support Putting AI Ethics into Practice. In Philosophy & Technology 35 (3), pp. 1–246. (Link)
• Hagendorff, Thilo; Danks, David (2022): Ethical and methodological challenges in building morally informed AI systems. In AI Ethics, pp. 1–14. (Link)
• Hagendorff, Thilo (2022): AI ethics and its pitfalls: not living up to its own standards? In AI and Ethics, pp. 1–8. (Link)
• Hagendorff, Thilo (2022): Blind spots in AI ethics. In AI Ethics 2 (4), pp. 851–867. (Link)
• Hagendorff, Thilo (2021): Linking human and machine behavior. A new approach to evaluate training data quality for beneficial machine learning. In Minds and Machines 31, pp. 563–593. (Link)
• Hagendorff, Thilo; Meding, Kristof (2021): Ethical considerations and statistical analysis of industry involvement in machine learning research. In AI & SOCIETY - Journal of Knowledge, Culture and Communication, pp. 1–11. (Link)
• Hagendorff, Thilo (2021): Forbidden knowledge in machine learning. Reflections on the limits of research and publication. In AI & SOCIETY - Journal of Knowledge, Culture and Communication 36 (3), pp. 767–781. (Link)
• Helm, Paula; Hagendorff, Thilo (2021): Beyond the Prediction Paradigm. Challenges for AI in the Struggle Against Organized Crime. In Law and Contemporary Problems 84 (3), pp. 1–17. (Link)
• Hagendorff, Thilo (2020): The Ethics of AI Ethics. An Evaluation of Guidelines. In: Minds and Machines 30 (3), pp. 457–461. (Link)
• Hagendorff, Thilo (2019): From privacy to anti-discrimination in times of machine learning. In: Ethics and Information Technology 33 (3), pp. 331–343. (Link)
• Hagendorff, Thilo; Wezel, Katharina (2019): 15 challenges for AI: or what AI (currently) can’t do. In AI & SOCIETY - Journal of Knowledge, Culture and Communication 35 (2), pp. 355-365. (Link)

Thilo Hagendorff leads the AI safety research group at the University of Stuttgart. He studies how AI agents behave, with experiments on deception, emergent misalignment, evaluation awareness, or shutdown resistance. His works appeared in PNAS, Science Advances, Nature Communications, or Nature Computational Science. More than 200 media reports have covered his research, including in The Economist, MIT Technology Review, and Scientific American. He has held visiting positions at Stanford, UC San Diego, ELLIS Alicante, and the University of Oslo. His group is funded by Coefficient Giving and Schmidt Sciences.

More details can be found here.