PeopleofSuperintelligence

Researcher

Akbir Khan

Akbir Khan is an AI researcher whose work spans language pragmatics, multi-agent learning, transfer learning, and scalable oversight. He studied mathematics and physics at UCL and computer science at Cambridge, co-founded Spherical Defence Labs, and completed a PhD at UCL DARK in 2025. His research includes a 2019 study of transfer learning, MAESTRO, a NeurIPS study of pragmatic understanding, and a 2024 ICML Best Paper on debate. In 2025, he co-authored a report on multi-agent risks; as of October 2026, he is a Member of Technical Staff at Anthropic.

Now at
Anthropic
Badges
Links
Google Scholar and GitHub

Timeline

  1. 2019

    Race and transfer learning

    In 2019, he co-authors Considering Race a Problem of Transfer Learning, which is accepted for an oral presentation at DVPBA. The paper examines race as a boundary for transfer learning in facial classification and image synthesis.

  2. 2021–2025

    Education and a UCL PhD

    He studies mathematics and physics at UCL and computer science at Cambridge, and co-founds Spherical Defence Labs. He pursues a PhD at UCL DARK from 2021 to 2025.

  3. 2023

    MAESTRO for multi-agent RL

    In 2023, he co-authors MAESTRO, a multi-agent environment-design method for open-ended reinforcement learning. The method builds joint curricula over environments and co-players for two-player zero-sum games.

  4. 2023

    Pragmatic understanding in LLMs

    At NeurIPS 2023, he co-authors The Goldilocks of Pragmatic Understanding. The study finds that example-level instruction-tuned LLMs perform significantly better on implicature tasks.

  5. 2024

    Opponent shaping at AAMAS

    At AAMAS 2024, he co-authors Scaling Opponent Shaping to High Dimensional Games. The paper scales an opponent-shaping method to general-sum games with temporally extended actions and long horizons.

  6. 2024

    An ICML debate paper

    In 2024, he co-authors Debating with More Persuasive LLMs Leads to More Truthful Answers. The paper reports 76% accuracy for non-expert models and 88% for humans, and ICML names it a Best Paper.

  7. 2025

    Multi-agent risks report

    In 2025, he co-authors Multi-Agent Risks from Advanced AI. The report identifies miscoordination, conflict, and collusion as three failure modes in multi-agent systems.

  8. As of October 2026

    Anthropic technical staff

    In March 2025, his research centers on scalable oversight through multi-agent learning. As of October 2026, he is a Member of Technical Staff at Anthropic.

Career

  1. NowAnthropicMember of Technical Staff
  2. Spherical Defence LabsCo-founder

Education

  1. 2025University College LondonPhD
  2. University of Cambridge

From the biography.

Videos

YouTube videos in which Akbir Khan speaks or appears, newest first. A model chose them from YouTube search results. List updated 21 October 2026.

  1. This is what happens when you let AIs debateMachine Learning Street Talk · 27 September 2024 · 46 min

Video data from YouTube API Services. By watching you agree to the YouTube Terms of Service. How we use YouTube data.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Akbir Khan. 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. 16% of AI career moves in the 2020s went from a university to a companyIncludes Akbir Khan. 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.
  3. 22 of 133 people shaping AI with a former OpenAI role now work at AnthropicIncludes Anthropic. Our estimates of where OpenAI recruits came from and where 133 people with former OpenAI roles work now, including the companies and labs they founded.
  4. Companies supply the most prior-work hires at 7 of the 9 frontier AI labsIncludes Anthropic. Our estimates of the organizations where 550 people worked directly before joining 9 frontier AI labs across 620 career transitions, comparing the main feeder organizations per lab.

Sources and credits

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

  1. Considering Race a Problem of Transfer Learning arxiv.org. Checked 21 October 2026.
  2. akbir khan akbir.dev. Checked 21 October 2026.
  3. UCL DARK Lab ucldark.com. Checked 21 October 2026.
  4. MAESTRO: Open-Ended Environment Design for Multi-Agent Reinforcement Learning arxiv.org. Checked 21 October 2026.
  5. The Goldilocks of Pragmatic Understanding: Fine-Tuning Strategy Matters for Implicature Resolution by LLMs arxiv.org. Checked 21 October 2026.
  6. Accepted Papers | The 23rd International Conference on Autonomous Agents and Multi-Agent Systems aamas2024-conference.auckland.ac.nz. Checked 21 October 2026.
  7. Scaling Opponent Shaping to High Dimensional Games arxiv.org. Checked 21 October 2026.
  8. Debating with More Persuasive LLMs Leads to More Truthful Answers proceedings.mlr.press. Checked 21 October 2026.
  9. ICML 2024 Awards | ICML 2024 icml.cc. Checked 21 October 2026.
  10. Multi-Agent Risks from Advanced AI arxiv.org. Checked 21 October 2026.
  11. The Stanford Natural Language Processing Group nlp.stanford.edu. Checked 21 October 2026.
  12. Cooperative AI – Foundation cooperativeai.com. Checked 21 October 2026.