# Beth Barnes

Founder and CEO, METR. TIME100 AI 2026, Thinkers. She left OpenAI to test AI from the outside; her group's chart shows AI's task length doubling every four to seven months. ([time.com](https://time.com/collection/time100-ai/2026/beth-barnes/))

- Education: Computer science, University of Cambridge, 2018 ([linkedin.com](https://www.linkedin.com/in/elizabethmbarnes))
- Career: METR (Founder and CEO, 2022 to now); OpenAI (Researcher, Alignment Team, 2019–2022); DeepMind (Research Assistant to Chief Scientist, 2018) ([metr.org](https://metr.org/about), [time.com](https://time.com/collection/time100-ai/2026/beth-barnes/), [linkedin.com](https://www.linkedin.com/in/elizabethmbarnes), [aisfoundation.ai](https://aisfoundation.ai/committee/beth-barnes))
- TIME profile: https://time.com/collection/time100-ai/2026/beth-barnes/
- Profile: https://www.autumn.ai/resources/most-influential-people-in-ai/beth-barnes

- Left OpenAI's alignment team in 2022 to build an outside group that tests AI models, which became the independent nonprofit METR. ([time.com](https://time.com/collection/time100-ai/2026/beth-barnes/))
- Her team found that the length of tasks AI can complete, measured in human time, has doubled every four to seven months since 2023. ([time.com](https://time.com/collection/time100-ai/2026/beth-barnes/))
- METR's 2025 study found that AI coding assistants actually slowed experienced engineers down. ([time.com](https://time.com/collection/time100-ai/2026/beth-barnes/))
- Her team ran dangerous-capability tests on GPT-4 and Claude before release as a third-party evaluator. ([alignmentforum.org](https://www.alignmentforum.org/users/beth-barnes))
- Told a podcast: "I am an expert telling you you should freak out." ([80000hours.org](https://80000hours.org/podcast/episodes/beth-barnes-ai-safety-evals/))