PeopleofSuperintelligence

Researcher

Daniel Ziegler

Also known as Daniel M. Ziegler

Daniel Ziegler is a machine-learning researcher who has worked on AI alignment, the effort to make AI systems behave as their designers intend. By 2020 he was a research engineer at OpenAI, having left a Stanford ML PhD, and he spent three years there building RLHF and GPT-3. He co-authored the GPT-3 paper, 'Language Models are Few-Shot Learners', which NeurIPS named an Outstanding Paper in 2020. At Redwood Research he led the adversarial training team in 2022, and in March 2025 he co-authored a study on auditing language models for hidden objectives. In October 2025 he was named as one of two internal reviewers of Anthropic's pilot sabotage risk report.

Timeline

  1. 2015

    Summer fellowship at MIRI

    In 2015 he attends the MIRI Summer Fellows program, after formal verification research at MIT.

  2. By 2020

    Research engineer at OpenAI

    By 2020 he is a research engineer at OpenAI, having left a Stanford ML PhD. There he spends three years building RLHF and GPT-3.

  3. 2019–2020

    Fine-tuning from human preferences

    The paper 'Fine-Tuning Language Models from Human Preferences' lists him as co-author and is posted to arXiv in September 2019. In September 2020 he co-authors 'Learning to summarize from human feedback'.

  4. 2020

    A NeurIPS Outstanding Paper

    GPT-3's paper, 'Language Models are Few-Shot Learners', lists him as co-author, and NeurIPS names it an Outstanding Paper in 2020.

  5. 2022

    Adversarial training at Redwood

    In 2022 he leads the adversarial training team at Redwood Research, after time on OpenAI's alignment team. His paper 'Adversarial Training for High-Stakes Reliability' is posted to arXiv in May 2022.

  6. January 2024

    Co-authoring Sleeper Agents

    He is a co-author of 'Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training', posted to arXiv in January 2024.

  7. March 2025

    Auditing hidden objectives

    He co-authors 'Auditing language models for hidden objectives', posted to arXiv in March 2025.

  8. October 2025

    Internal review of a risk report

    In October 2025 he is named as one of two internal reviewers of Anthropic's pilot sabotage risk report.

Sources and credits

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

  1. Daniel Ziegler | Lynette Bye lynettebye.com. Checked 14 October 2026.
  2. Daniel Ziegler (former) - METR metr.org. Checked 14 October 2026.
  3. PhD or programming? Fast paths into aligning AI as a machine learning engineer, according to ML engineers Catherine Olsson & Daniel Ziegler | 80,000 Hours 80000hours.org. Checked 14 October 2026.
  4. [1909.08593] Fine-Tuning Language Models from Human Preferences arxiv.org. Checked 14 October 2026.
  5. [2009.01325] Learning to summarize from human feedback arxiv.org. Checked 14 October 2026.
  6. NeurIPS 2020 Awards neurips.cc. Checked 14 October 2026.
  7. [2005.14165] Language Models are Few-Shot Learners arxiv.org. Checked 14 October 2026.
  8. 17 - Training for Very High Reliability with Daniel Ziegler | AXRP - the AI X-risk Research Podcast axrp.net. Checked 14 October 2026.
  9. [2205.01663] Adversarial Training for High-Stakes Reliability arxiv.org. Checked 14 October 2026.
  10. [2401.05566] Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training arxiv.org. Checked 14 October 2026.
  11. [2503.10965] Auditing language models for hidden objectives arxiv.org. Checked 14 October 2026.
  12. Anthropic's Pilot Sabotage Risk Report alignment.anthropic.com. Checked 14 October 2026.