PeopleofSuperintelligence

Researcher

Owain Evans

Owain Evans is an AI safety researcher whose work tests how systems infer human values, answer truthfully, and behave after training. He has a PhD from MIT and, at Oxford's Future of Humanity Institute, works on algorithms that learn human preferences from imperfect choices. His research includes a survey of expert AI forecasts, a paper on standards for truthful AI, and the TruthfulQA benchmark. After advising Apollo Research, he gives the 2025 Hinton Lectures on AI safety and coauthors a Nature study on emergent misalignment. As of October 2026, he directs Truthful AI, serves on Ought's board, and appears on TIME's TIME100 AI list.

Now at
Truthful AI, University of California, Berkeley and Ought
Works
45
Citations
3,695
h-index
14

Publication numbers from OpenAlex, as of 14 October 2026.

Photo of Owain Evans

Timeline

  1. 2016

    Learning human preferences

    He has a PhD from MIT and is an FHI-Amlin postdoctoral research fellow at Oxford's Future of Humanity Institute. His AI safety project aims to develop algorithms that learn human preferences from choices shaped by bias and inconsistency.

  2. 2017–2018

    Forecasting AI progress

    With four coauthors, he reports a large survey of machine-learning researchers' beliefs about AI progress, including when they expect AI to outperform humans.

  3. 2021

    Defining truthful AI

    In “Truthful AI: Developing and governing AI that does not lie,” he and seven coauthors propose standards, evaluation institutions, and training methods for robustly truthful AI systems.

  4. 2022

    Testing model truthfulness

    With Stephanie Lin and Jacob Hilton, he publishes TruthfulQA, an 817-question benchmark that tests whether language models give truthful answers and mimic common misconceptions.

  5. April 2024

    Advising Apollo Research

    Apollo Research lists him among three official advisers who regularly meet with the team. The lab works on evaluations for deceptive alignment and scheming.

  6. 2025

    Teaching AI safety

    In Toronto, he gives the 2025 Hinton Lectures, a three-part AI safety series for a general audience hosted by Geoffrey Hinton.

  7. January 2026

    Finding broad misalignment

    He coauthors a Nature study on emergent misalignment, showing that narrow fine-tuning on insecure code can lead language models to give malicious advice and deceptive answers.

  8. As of October 2026

    Leading Truthful AI

    He directs Truthful AI and is an affiliate at CHAI, UC Berkeley; he also remains on the board of Ought after working there. TIME's 2026 TIME100 AI list profiles his experiments that make AI safety concerns testable.

Sources and credits

Photo: owainevans.github.io; Photos of Owain Evans; creator not stated. License: Not stated by source. Source: owainevans.github.io. Cropped to a square and resized. Ask us to remove or replace this photo.

  1. AI Researcher Owain Evans - Future of Life Institute futureoflife.org. Checked 14 October 2026.
  2. Team | Truthful AI truthful.ai. Checked 14 October 2026.
  3. [1705.08807] When Will AI Exceed Human Performance? Evidence from AI Experts arxiv.org. Checked 14 October 2026.
  4. [2110.06674] Truthful AI: Developing and governing AI that does not lie arxiv.org. Checked 14 October 2026.
  5. TruthfulQA: Measuring How Models Mimic Human Falsehoods - ACL Anthology aclanthology.org. Checked 14 October 2026.
  6. The First Year of Apollo Research apolloresearch.ai. Checked 14 October 2026.
  7. Owain Evans, AI Alignment researcher owainevans.github.io. Checked 14 October 2026.
  8. Training large language models on narrow tasks can lead to broad misalignment | Nature nature.com. Checked 14 October 2026.
  9. Owain Evans: The 100 Most Influential People in AI 2026 time.com. Checked 14 October 2026.