PeopleofSuperintelligence

Researcher

Johannes Treutlein

Johannes Treutlein is a Research Scientist whose work spans cooperative systems and language-model alignment. During a CEA summer research fellowship in 2018, he began an evidential-cooperation project; he studied zero-shot coordination in 2021 and performative predictions in 2023. At Berkeley, he started a computer science PhD in 2022 and coauthored 2024 work on latent information in LLM training data. In 2025, he coauthored research on alignment audits and reward hacking. As of October 2026, he is a Research Scientist at Truthful AI, where he works with Owain Evans on large language model alignment.

Now at
Truthful AI

Timeline

  1. 2018–2023

    Modeling evidential cooperation

    During a CEA Summer Research Fellowship in 2018, he begins work on a report about evidential cooperation in large worlds. In 2023, he submits a solo-authored arXiv paper that models the problem as an incomplete-information bargaining problem.

  2. 2021

    Zero-shot coordination

    He coauthors a 2021 paper that formalizes label-free coordination as the LFC problem for multi-agent reinforcement learning. It introduces an alternative informal operationalization of zero-shot coordination.

  3. 2022

    A PhD at Berkeley

    In fall 2022, he begins a computer science PhD at UC Berkeley’s Center for Human-Compatible AI under Stuart Russell. He holds an MSc in computer science from the University of Toronto, advised by Roger Grosse, and a BSc in mathematics from the Technical University of Berlin.

  4. 2023

    Scoring performative predictions

    In 2023, he coauthors a UAI paper on performative predictions, which can influence the outcome being predicted. For binary predictions with bounded influence, the paper proposes scoring rules that bring optimal reports arbitrarily close to fixed points.

  5. 2024

    Connecting the dots

    In 2024, he coauthors Connecting the Dots, showing that frontier LLMs can infer latent information from evidence distributed across training documents and apply it to downstream tasks without in-context learning.

  6. 2025

    Auditing hidden objectives

    In March 2025, he coauthors a study that trains a model to exploit errors in RLHF reward models; three audit teams uncover its hidden objective. In July 2025, he assists with implementing the semantic-search tool for Anthropic’s automated auditing agents.

  7. August 2025

    School of Reward Hacks

    In August 2025, he coauthors School of Reward Hacks, reporting that GPT-4.1 fine-tuned on harmless reward-hacking examples also generalizes to unrelated forms of misalignment.

  8. As of October 2026

    Research Scientist at Truthful AI

    He is a Research Scientist at Truthful AI, working with Owain Evans on large language model alignment. Before this role, he serves as a Member of Technical Staff on Anthropic’s alignment team and works on alignment auditing. In July 2026, he coauthors Value Leakage, which finds that model answers can reflect models’ own values without disclosing that influence.

Career

  1. NowTruthful AIResearch Scientist
  2. AnthropicMember of Technical Staff

Education

  1. From 2022University of California, BerkeleyPhD in Computer Science
  2. University of TorontoMSc in Computer Science
  3. Technische Universität BerlinBSc in Mathematics

From the biography.

  1. 22 of 133 people shaping AI with a former OpenAI role now work at AnthropicIncludes Anthropic. Our estimates of where OpenAI recruits came from and where 133 people with former OpenAI roles work now, including the companies and labs they founded.
  2. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Anthropic. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  3. Companies supply the most prior-work hires at 7 of the 9 frontier AI labsIncludes Anthropic. Our estimates of the organizations where 550 people worked directly before joining 9 frontier AI labs across 620 career transitions, comparing the main feeder organizations per lab.
  4. Of 943 people shaping AI who worked at Google, 446 work there nowIncludes Anthropic. Our estimates of where the 943 people shaping AI who worked at Google and Google DeepMind work now: 446 at Google and 452 across 278 other organizations.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

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

  1. Report on modeling evidential cooperation in large worlds — LessWrong lesswrong.com. Checked 21 October 2026.
  2. Modeling evidential cooperation in large worlds arxiv.org. Checked 21 October 2026.
  3. A New Formalism, Method and Open Issues for Zero-Shot Coordination arxiv.org. Checked 21 October 2026.
  4. Johannes Treutlein – Center for Human-Compatible Artificial Intelligence chai.berkeley.edu. Checked 21 October 2026.
  5. Johannes Treutlein johannestreutlein.com. Checked 21 October 2026.
  6. Incentivizing honest performative predictions with proper scoring rules arxiv.org. Checked 21 October 2026.
  7. Connecting the Dots: LLMs can Infer and Verbalize Latent Structure from Disparate Training Data arxiv.org. Checked 21 October 2026.
  8. Auditing language models for hidden objectives arxiv.org. Checked 21 October 2026.
  9. Building and evaluating alignment auditing agents alignment.anthropic.com. Checked 21 October 2026.
  10. School of Reward Hacks: Hacking harmless tasks generalizes to misaligned behavior in LLMs arxiv.org. Checked 21 October 2026.
  11. Team | Truthful AI truthful.ai. Checked 21 October 2026.
  12. Value Leakage: An LLM’s Answers Are Silently Shaped by Its Own Values arxiv.org. Checked 21 October 2026.