# Ethan Perez

Research and product · Alignment research lead at Anthropic · joined 2022

Co-wrote the original retrieval-augmented generation paper as a PhD student, then moved to alignment research.

Full profile: https://www.autumn.ai/resources/anthropic/people/ethan-perez

**At Anthropic:**
- 2024 – present: Alignment team lead
- 2022: Research Scientist; led the adversarial robustness team
Sources: [1](https://ethanperez.net/) [2](https://www.alphaxiv.org/@ethan-perez)

**Education:**
- PhD, Natural Language Processing/Computer Science, New York University
- Bachelor's, Rice University (Engineering department's Outstanding Senior)
Sources: [1](https://www.far.ai/author/ethan-perez)

**Lesser-known facts:**
- Considered quitting an internship to start a robotics company with a friend. [1](https://80000hours.org/stories/ethan-perez/)
- Won a best paper award at ICML 2024 and made Forbes 30 Under 30 in AI. [1](https://ethanperez.net/)
- Discovered the career group 80,000 Hours as a computer science undergraduate, then interned at Mila in Quebec. [1](https://80000hours.org/stories/ethan-perez/)
- Attended a Machine Intelligence Research Institute workshop on AI safety in 2017, while still at Rice. [1](https://ethanperez.net/wp-content/uploads/2020/11/EthanPerezCurriculumVitae.pdf)
- Interned at Uber on fraud detection through a KPCB Fellowship and at Google Maps on localization algorithms. [1](https://ethanperez.net/wp-content/uploads/2020/11/EthanPerezCurriculumVitae.pdf)
- Founded and coached a MathCounts program at St. Mark's Catholic School in 2012. [1](https://ethanperez.net/wp-content/uploads/2020/11/EthanPerezCurriculumVitae.pdf)
- Wrote his NYU PhD thesis on fixing undesirable behavior in language models. [1](https://80000hours.org/stories/ethan-perez/)
- Led the team behind Constitutional Classifiers, which let Anthropic deploy Claude Opus 4 despite its ability to assist advanced weapons development. [1](https://ethanperez.net/)

**Known for:**
- Retrieval-Augmented Generation (RAG, 2020)
- Automated Red Teaming (2022)
- Constitutional Classifiers
- Sleeper Agents (2024)
- Measuring faithfulness in chain-of-thought reasoning (2023)
- Question Decomposition Improves the Faithfulness of Model-Generated Reasoning (2023)
Sources: [1](https://ethanperez.net/)

Researched with Autumn (https://www.autumn.ai) from public sources. Updated 2026-10-18. Everyone at Anthropic: https://www.autumn.ai/resources/anthropic/people