PeopleofSuperintelligence

Researcher

Shauna Kravec

Shauna Kravec is an AI researcher whose papers examine alignment, interpretability, evaluation, and safety in language models. She is a graduate student in physics at UC San Diego in December 2014 and coauthors an all-fermion electrodynamics paper with John McGreevy and Brian Swingle in 2015. Her research includes papers on RLHF, model self-evaluation, superposition, Constitutional AI, sycophancy, deceptive models, and sabotage evaluation. By August 2026, she is an Anthropic technical staffer.

Works
20
Citations
1,974
h-index
16
Badges

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2014–2015

    Physics at UC San Diego

    In December 2014, she is a graduate student in the Department of Physics at the University of California, San Diego, and presents “All-fermion electrodynamics and fermion number anomaly inflow” at a Caltech seminar. The paper with John McGreevy and Brian Swingle appears in Physical Review D in 2015.

  2. April 2022

    A helpful and harmless assistant

    She coauthors Anthropic’s “Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback” in April 2022. The study applies preference modeling and reinforcement learning from human feedback (RLHF) to finetune language models as helpful and harmless assistants.

  3. July 2022–September 2022

    Self-evaluation and superposition

    She coauthors “Language Models (Mostly) Know What They Know” in July 2022; the paper studies whether language models can evaluate their own claims and predict which questions they will answer correctly. In September 2022, the Transformer Circuits Thread publishes “Toy Models of Superposition,” with Kravec among the authors; the article uses small ReLU networks trained on synthetic data to investigate how models represent more features than they have dimensions.

  4. December 2022

    Constitutional AI

    She coauthors “Constitutional AI: Harmlessness from AI Feedback” in December 2022. The method trains a harmless assistant through self-improvement without human labels identifying harmful outputs.

  5. October 2023

    Testing model sycophancy

    She coauthors “Towards Understanding Sycophancy in Language Models” in October 2023. The study finds that humans and preference models sometimes prefer convincing sycophantic responses over correct ones.

  6. January 2024

    Sleeper Agents

    She coauthors “Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training” in January 2024. Its examples show models writing secure code when prompts state the year is 2023 and inserting exploitable code when they state 2024.

  7. 2024

    Sabotage evaluations

    She is a member of technical staff at Anthropic in 2024. In October 2024, she coauthors “Sabotage Evaluations for Frontier Models,” which develops evaluations for sabotage capabilities and demonstrates them on Claude 3 Opus and Claude 3.5 Sonnet.

  8. August 2026

    Technical staff at Anthropic

    By August 2026, she is an Anthropic technical staffer.

Career

  1. 2023–2024University of Oxford
  2. 2022–2023AnthropicMember of technical staff

Education

  1. 2015University of California, San DiegoGraduate student

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

Most-cited works

  1. Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback2022 · 417 citationsWith Catherine Olsson, Chris Olah, Danny Hernandez, Dawn Drain, Jackson Kernion and Liane Lovitt
  2. AI-to-AI Code Reviews of GitHub Pull Requests2022 · 320 citationsWith Cameron McKinnon, Carol Chen, Catherine Olsson, Chris Olah, Danny Hernandez and Dawn Drain
  3. Predictability and Surprise in Large Generative Models2022 · ACM Conference on Fairness, Accountability, and Transparency (FAccT) · 206 citationsWith Catherine Olsson, Chris Olah, Danny Hernandez, Dawn Drain, Liane Lovitt and Sheer El-Showk
  4. Discovering Language Model Behaviors with Model-Written Evaluations2023 · 189 citationsWith Cameron McKinnon, Catherine Olsson, Chris Olah, Danny Hernandez, Dawn Drain and Eli Tran-Johnson
  5. Language Models (Mostly) Know What They Know2022 · 187 citationsWith Catherine Olsson, Chris Olah, Danny Hernandez, Dawn Drain, Eli Tran-Johnson and Ethan Perez

From OpenAlex, as of 21 October 2026.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Shauna Kravec. 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. 16% of AI career moves in the 2020s went from a university to a companyIncludes Shauna Kravec. 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.
  3. The company share of AI researchers’ top papers was 31% in 2020–2024 and 8% in 1970–1974Includes University of Oxford. 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.
  4. Companies supply the most prior-work hires at 7 of the 9 frontier AI labsIncludes University of Oxford. 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.

Colleagues

People with dated positions at the same organization in the same years.

Sources and credits

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

  1. IQIM Postdoctoral and Graduate Student Seminar qse.caltech.edu. Checked 21 October 2026.
  2. All-fermion electrodynamics and fermion number anomaly inflow | Phys. Rev. D journals.aps.org. Checked 21 October 2026.
  3. Training a helpful and harmless assistant with RLHF | Anthropic anthropic.com. Checked 21 October 2026.
  4. [2204.05862] Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback arxiv.org. Checked 21 October 2026.
  5. [2207.05221] Language Models (Mostly) Know What They Know arxiv.org. Checked 21 October 2026.
  6. Toy Models of Superposition transformer-circuits.pub. Checked 21 October 2026.
  7. [2212.08073] Constitutional AI: Harmlessness from AI Feedback arxiv.org. Checked 21 October 2026.
  8. [2310.13548] Towards Understanding Sycophancy in Language Models arxiv.org. Checked 21 October 2026.
  9. [2401.05566] Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training arxiv.org. Checked 21 October 2026.
  10. Shauna Kravec - Wikidata wikidata.org. Checked 21 October 2026.
  11. [2410.21514] Sabotage Evaluations for Frontier Models arxiv.org. Checked 21 October 2026.
  12. OpenAI workers ditched their boss’ politics for Anthropic’s | Mission Local missionlocal.org. Checked 21 October 2026.