PeopleofSuperintelligence

Researcher

Jesse Mu

Jesse Mu is a member of technical staff at Anthropic whose research uses language to train and interpret machine learning models. After studying computer science at Boston College, he earned a Cambridge MPhil on a Churchill Scholarship and completed a Stanford PhD advised by Noah Goodman in 2023. His papers include Compositional Explanations of Neurons, written with Jacob Andreas in 2020, and a 2023 method that compresses language model prompts into gist tokens. In 2024, he coauthored Sleeper Agents, which showed that backdoor behavior in large language models can persist through standard safety training. He also coauthored Constitutional Classifiers, a 2025 set of safeguards against universal jailbreaks.

Now at
Anthropic
Badges

Timeline

  1. 2016–2017

    Computer science at Boston College

    From Omaha, Nebraska, he majors in computer science at Boston College and works with psychologist Joshua Hartshorne on computational modeling of human language, first at MIT. He wins a Barry M. Goldwater Scholarship in 2016 and graduates with a BA in 2017.

  2. 2017–2018

    Churchill Scholar at Cambridge

    In January 2017, he wins a Winston Churchill Scholarship, the first Boston College student to receive it since 2006. The scholarship funds his MPhil in Advanced Computer Science at the University of Cambridge, which he finishes in 2018.

  3. 2018–2023

    A Stanford PhD with Noah Goodman

    He starts a computer science PhD at Stanford in fall 2018 with an NSF Graduate Research Fellowship, planning work on natural language processing, psycholinguistics and machine learning. During the PhD, he spends time at MIT LINGO, FAIR and DeepMind; advised by Noah Goodman and also supported by an Open Phil AI Fellowship, he completes the degree in 2023.

  4. 2020

    Learning from language explanations

    At ACL 2020, he presents Shaping Visual Representations with Language for Few-shot Classification with Percy Liang and Noah Goodman. In November 2020, he and Shikhar Murty write a Stanford AI Lab blog post on deep neural models that learn from language explanations in NLP and computer vision.

  5. 2020

    Compositional explanations of neurons

    With Jacob Andreas, he publishes Compositional Explanations of Neurons as an oral paper at NeurIPS 2020. The method explains neurons in vision and language models with compositional logical concepts and finds polysemantic neurons that detect multiple unrelated features.

  6. 2023

    Gist tokens for prompt compression

    With Xiang Lisa Li and Noah Goodman, he presents gisting at NeurIPS 2023, a method that trains a language model to compress prompts into smaller sets of gist tokens that can be cached and reused. On LLaMA-7B and FLAN-T5-XXL, it compresses prompts up to 26 times with minimal loss in output quality.

  7. 2024

    The Sleeper Agents study

    In January 2024, he coauthors Sleeper Agents with Evan Hubinger, Ethan Perez and others, training large language models that write secure code when a prompt says the year is 2023 and insert exploitable code when it says 2024. This backdoor behavior can persist through standard safety training, including supervised fine-tuning, reinforcement learning and adversarial training. He also coauthors Many-shot Jailbreaking, presented at NeurIPS 2024.

  8. As of October 2026

    Member of technical staff at Anthropic

    He is a member of technical staff at Anthropic, with research interests in using language to train and interpret machine learning models. In January 2025, he coauthors Constitutional Classifiers, safeguards trained on synthetic data generated from a constitution of natural-language rules; in over 3,000 estimated hours of red teaming, no red teamer finds a universal jailbreak of an early classifier-guarded model.

Career

  1. NowAnthropicMember of technical staff

Education

  1. 2023Stanford UniversityPhD
  2. 2018University of CambridgeMPhil in Advanced Computer Science
  3. 2017Boston CollegeBA

From the biography.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Jesse Mu. 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.
  2. Of 943 people shaping AI who worked at Google, 446 work there nowIncludes Jesse Mu. 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.
  3. 16% of AI career moves in the 2020s went from a university to a companyIncludes Jesse Mu. We estimate that 16% of AI career moves in the 2020s went from a university to a company; 29% of moves starting at a university went to a company, and 76% ended the university role.
  4. 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.

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 Jesse Mu yet. Send us a free-licensed photo.

  1. BC Senior Jesse Mu Wins Prestigious Churchill Scholarship bc.edu. Checked 20 October 2026.
  2. Seniors to Remember: Jesse Mu bc.edu. Checked 20 October 2026.
  3. Jesse Mu jesse.mu. Checked 20 October 2026.
  4. Jesse Mu Awarded NSF GRFP Fellowship bc.edu. Checked 20 October 2026.
  5. Learning from Language Explanations | SAIL Blog ai.stanford.edu. Checked 20 October 2026.
  6. Compositional Explanations of Neurons arxiv.org. Checked 20 October 2026.
  7. Learning to Compress Prompts with Gist Tokens arxiv.org. Checked 20 October 2026.
  8. Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training arxiv.org. Checked 20 October 2026.
  9. Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teaming arxiv.org. Checked 20 October 2026.