PeopleofSuperintelligence

Researcher

Dawn Drain

Dawn Drain is a researcher whose work spans code generation and language-model alignment. She co-authored GraphCodeBERT in 2020 and contributed to Microsoft's 2021 work on generating bug fixes with pretrained transformers. Her research then covered language assistants, reinforcement learning from human feedback, Constitutional AI, and models' ability to assess their own answers. In 2025, she helped implement SWE-bench and train Claude 3.5 Sonnet for agentic coding. As of October 2026, she is a research engineer at Anthropic.

Now at
Anthropic
Works
28
Citations
3,234
h-index
20
Badges

Publication numbers from OpenAlex, as of 19 October 2026.

Timeline

  1. 2020

    GraphCodeBERT and code

    In 2020, she co-authors GraphCodeBERT, a pretrained language model that uses data flow to represent code structure.

  2. 2021

    Transformer bug fixes

    In 2021, she is affiliated with Microsoft and contributes to MAPS work on generating bug fixes with pretrained transformers.

  3. 2021

    A laboratory for alignment

    In 2021, she co-authors a paper that treats a general language assistant as a laboratory for alignment research.

  4. 2022

    Training helpful assistants

    In 2022, she co-authors Anthropic research on using preference modeling and reinforcement learning from human feedback to train helpful and harmless assistants.

  5. 2022

    Constitutional AI

    In 2022, she co-authors Constitutional AI, which trains a harmless assistant through self-improvement without human labels identifying harmful outputs.

  6. 2022

    Models assessing their knowledge

    In 2022, she co-authors research that tests whether language models can judge their own claims and predict which questions they can answer correctly.

  7. 2025

    An agentic coding benchmark

    In 2025, she helps implement the SWE-bench benchmark and contributes to training Claude 3.5 Sonnet for agentic coding.

  8. As of October 2026

    Research engineer at Anthropic

    As of October 2026, she is a research engineer at Anthropic. In January 2026, she is thanked for helpful discussions on pre-deployment auditing of models trained to sabotage Anthropic.

Career

  1. Since 2026AnthropicResearch engineer
  2. 2022–2023Machine Intelligence Research Institute
  3. 2021Microsoft

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

Most-cited works

  1. CodeXGLUE: A Machine Learning Benchmark Dataset for Code Understanding and Generation2021 · 536 citationsWith Daya Guo, Ming Zhou and Nan Duan
  2. GraphCodeBERT: Pre-training Code Representations with Data Flow2020 · 478 citationsWith Daya Guo, Ming Zhou and Nan Duan
  3. Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback2022 · 417 citationsWith Catherine Olsson, Chris Olah, Danny Hernandez, Liane Lovitt and Zac Hatfield-Dodds
  4. AI-to-AI Code Reviews of GitHub Pull Requests2022 · 319 citationsWith Cameron McKinnon, Carol Chen, Catherine Olsson, Chris Olah, Danny Hernandez and Ethan Perez
  5. Predictability and Surprise in Large Generative Models2022 · ACM Conference on Fairness, Accountability, and Transparency (FAccT) · 206 citationsWith Catherine Olsson, Chris Olah, Danny Hernandez, Liane Lovitt and Zac Hatfield-Dodds

From OpenAlex, as of 19 October 2026.

  1. The median h-index is 53 at universities and 19 at frontier AI labsIncludes Dawn Drain. The median h-index is 53 at universities and 19 at frontier AI labs. We compare publishing careers, citations and moves to estimate how universities and frontier labs contribute to AI research.
  2. Google alumni founded or led 87 other organizations, the most among the labs we compareIncludes Dawn Drain. We trace 212 organizations founded or led by AI lab alumni, and 223 career moves between the labs.
  3. 13 of 79 people shaping AI with a former OpenAI role now work at AnthropicIncludes Anthropic. Our estimates of where OpenAI recruits came from and where people with former OpenAI roles work now.
  4. Nonprofit roles are 2.2 times as common in AI safetyIncludes Anthropic. We estimate where people focused on safety, alignment and evaluation work. Nonprofit roles appear for 26.0% of safety people with known affiliations, compared with 11.6% in other areas.

Colleagues

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

Sources and credits

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

  1. GraphCodeBERT: Pre-training Code Representations with Data Flow microsoft.com. Checked 19 October 2026.
  2. Dawn Drain - HOPL IV hopl4.sigplan.org. Checked 19 October 2026.
  3. [2112.00861] A General Language Assistant as a Laboratory for Alignment arxiv.org. Checked 19 October 2026.
  4. Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback anthropic.com. Checked 19 October 2026.
  5. [2212.08073] Constitutional AI: Harmlessness from AI Feedback arxiv.org. Checked 19 October 2026.
  6. [2207.05221] Language Models (Mostly) Know What They Know arxiv.org. Checked 19 October 2026.
  7. Raising the bar on SWE-bench Verified with Claude 3.5 Sonnet anthropic.com. Checked 19 October 2026.
  8. Pre-deployment auditing can catch an overt saboteur alignment.anthropic.com. Checked 19 October 2026.
  9. Dawn Drain | 500AI 500ai.org. Checked 19 October 2026.