PeopleofSuperintelligence

Researcher

Sandhini Agarwal

As of March 2026, Sandhini Agarwal is an AI policy researcher on OpenAI's Deployment Planning team. She co-wrote the 2020 paper "Language Models are Few-Shot Learners," which NeurIPS named an Outstanding Paper. In 2021 she was first author of the "Evaluating CLIP" study, and in 2022 she co-wrote the paper on training language models with human feedback. In April 2025 she chaired OpenAI's Safety Advisory Group, the committee under which the company consolidated its safety governance.

Now at
OpenAI
Works
13

Publication numbers from OpenAlex, as of 15 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.

Timeline

  1. 2020

    A NeurIPS Outstanding Paper

    She co-writes "Language Models are Few-Shot Learners", a paper that NeurIPS names an Outstanding Paper in 2020.

  2. February 2021

    Visual models from language supervision

    She co-writes "Learning Transferable Visual Models From Natural Language Supervision", which is submitted to arXiv in February 2021.

  3. August 2021

    First author of Evaluating CLIP

    She is listed first of six authors on "Evaluating CLIP: Towards Characterization of Broader Capabilities and Downstream Implications", submitted to arXiv in August 2021.

  4. March 2022

    Instruction-following with human feedback

    She co-writes "Training language models to follow instructions with human feedback", which is submitted to arXiv in March 2022.

  5. April 2025

    Chairing the Safety Advisory Group

    A five-year OpenAI veteran, she chairs the Safety Advisory Group, under which the company consolidates its safety governance.

  6. March 2026

    AI policy research at OpenAI

    She works as an AI policy researcher on OpenAI's Deployment Planning team. Her earlier OpenAI work includes the bias analysis for GPT-3 and OpenAI's first content filter. In March 2026 she gives a talk at Caltech on collective alignment and public input on OpenAI's Model Spec.

Sources and credits

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

  1. [2005.14165] Language Models are Few-Shot Learners arxiv.org. Checked 15 October 2026.
  2. NeurIPS 2020 Awards neurips.cc. Checked 15 October 2026.
  3. [2103.00020] Learning Transferable Visual Models From Natural Language Supervision arxiv.org. Checked 15 October 2026.
  4. [2108.02818] Evaluating CLIP: Towards Characterization of Broader Capabilities and Downstream Implications arxiv.org. Checked 15 October 2026.
  5. [2203.02155] Training language models to follow instructions with human feedback arxiv.org. Checked 15 October 2026.
  6. Breaking: Top OpenAI Catastrophic Risk Official Steps Down Abruptly obsolete.pub. Checked 15 October 2026.
  7. CAIA Speaker Event: Sandhini Agarwal, OpenAI @ Caltech caltech.edu. Checked 15 October 2026.