PeopleofSuperintelligence

Researcher

Chawin Sitawarin

Also known as ชวิน สีตวาริน

Chawin Sitawarin is a machine-learning security researcher whose work spans adversarial robustness, Gemini security, and language-model privacy. He coauthored early research on deceptive traffic signs and later studies of robustness defenses and part-based models. At UC Berkeley, he completed a PhD in 2024 on adversarially robust machine-learning systems. His work on language-model memorization earned an ICML 2026 Outstanding Paper Honorable Mention, and he is a Member of Technical Staff at Anthropic on the Safeguards Research team as of October 2026.

Now at
Anthropic
Badges

Timeline

  1. 2018

    DARTS and toxic signs

    In 2018, he coauthors DARTS, which develops toxic-sign attacks against traffic-sign recognition in autonomous cars. The paper tests its attacks in virtual and real-world settings.

  2. 2022

    Random-transformation defenses

    In 2022, he coauthors Demystifying the Adversarial Robustness of Random Transformation Defenses. The study's attack reduces accuracy on Imagenette by 83%, compared with 19% for the commonly used EoT attack.

  3. 2023

    Part-based robustness models

    In 2023, he coauthors Part-Based Models Improve Adversarial Robustness, which combines part segmentation with a small classifier. The model reaches higher accuracy and adversarial robustness than a ResNet-50 baseline on three datasets.

  4. 2024

    A Berkeley security PhD

    He completes a PhD in UC Berkeley's EECS Department in 2024 under adviser David A. Wagner. His dissertation examines machine-learning defenses and presents a low-cost attack against a real-world language-model API.

  5. 2025

    Meta and Google DeepMind

    His work moves from a postdoctoral role in Meta's Privacy-Preserving Machine Learning team to a research scientist role in Google DeepMind's Security and Privacy team. In 2025, he coauthors a report on Google DeepMind's adaptive evaluations of Gemini against indirect prompt injection.

  6. 2025

    Measuring language-model memorization

    In 2025, he coauthors How much do language models memorize?, which estimates GPT-style models' capacity at about 3.6 bits per parameter. The paper distinguishes dataset-specific memorization from generalization.

  7. 2026

    An ICML honorable mention

    In 2026, How much can language models memorize? receives an ICML Outstanding Paper Honorable Mention, and he is one of its eight authors.

  8. As of October 2026

    Anthropic safeguards research

    As of October 2026, he is a Member of Technical Staff at Anthropic on the Safeguards Research team.

Career

  1. NowAnthropicMember of Technical Staff
  2. Until 2025MetaPostdoctoral Researcher
  3. From 2025Google DeepMindResearch Scientist

Education

  1. 2024University of California, BerkeleyPhD

From the biography.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Chawin Sitawarin. 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 Chawin Sitawarin. 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. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Chawin Sitawarin. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.
  4. 16% of AI career moves in the 2020s went from a university to a companyIncludes Chawin Sitawarin. 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.

Colleagues

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

Sources and credits

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

  1. DARTS: Deceiving Autonomous Cars with Toxic Signs arxiv.org. Checked 21 October 2026.
  2. Demystifying the Adversarial Robustness of Random Transformation Defenses arxiv.org. Checked 21 October 2026.
  3. Part-Based Models Improve Adversarial Robustness arxiv.org. Checked 21 October 2026.
  4. New Perspectives on Adversarially Robust Machine Learning Systems www2.eecs.berkeley.edu. Checked 21 October 2026.
  5. Chawin Sitawarin chawin.sitawarin.com. Checked 21 October 2026.
  6. Lessons from Defending Gemini Against Indirect Prompt Injections arxiv.org. Checked 21 October 2026.
  7. How much do language models memorize? arxiv.org. Checked 21 October 2026.
  8. ICML 2026 Awards icml.cc. Checked 21 October 2026.