PeopleofSuperintelligence

Researcher

Samuel Marks

Also known as Sam Marks

Samuel Marks is an AI researcher whose work spans interpretability and alignment audits of language models. He received a PhD from Harvard in May 2023 and then worked as a postdoc under David Bau in interpretability. With Max Tegmark, he posted The Geometry of Truth in October 2023, a study of how language models represent true and false statements. In March 2024 he posted Sparse Feature Circuits, and in March 2025 he posted Auditing language models for hidden objectives. He leads Anthropic's Cognitive Oversight team, which works on honesty, alignment auditing and improving oversight.

Now at
Anthropic
Works
9
Citations
24
h-index
2

Publication numbers from OpenAlex, as of 15 October 2026.

Timeline

  1. May 2023

    A Harvard PhD

    He receives a PhD from Harvard in May 2023. He then works as a postdoc under David Bau in interpretability.

  2. October 2023

    The geometry of truth

    In October 2023, he and Max Tegmark post The Geometry of Truth, a study of how language models represent true and false statements.

  3. March 2024

    Sparse feature circuits

    In March 2024, he posts Sparse Feature Circuits, a set of methods for finding human-interpretable features that explain language model behaviors.

  4. March 2025

    Auditing language models for hidden objectives

    In March 2025, he posts Auditing language models for hidden objectives, a study of alignment audits on a model trained with a hidden objective.

  5. July 2025

    Building and evaluating alignment auditing agents

    In July 2025, he and Evan Hubinger publish a post on alignment auditing agents.

  6. As of October 2026

    Cognitive oversight team

    He leads Anthropic's Cognitive Oversight team, a safety research team that works on honesty, alignment auditing, and improving oversight.

Sources and credits

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

  1. Samuel Marks people.math.harvard.edu. Checked 15 October 2026.
  2. Sam Marks - MATS Program Alumni matsprogram.org. Checked 15 October 2026.
  3. [2310.06824] The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets arxiv.org. Checked 15 October 2026.
  4. [2403.19647] Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models arxiv.org. Checked 15 October 2026.
  5. Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models arxiv.org. Checked 15 October 2026.
  6. [2503.10965] Auditing language models for hidden objectives arxiv.org. Checked 15 October 2026.
  7. Auditing language models for hidden objectives arxiv.org. Checked 15 October 2026.
  8. Building and evaluating alignment auditing agents alignment.anthropic.com. Checked 15 October 2026.