The Centre for the Governance of AI (GovAI) is an independent AI governance research nonprofit based in Oxford, United Kingdom. It produces conceptual and empirical research on compute governance, corporate governance at frontier labs, safety cases, and international coordination, and is the publisher of Safety Cases for Frontier AI.
| Field | Value |
|---|---|
| Type | Independent research nonprofit |
| Headquarters | Oxford, United Kingdom |
| Founded | 2018 (within the Future of Humanity Institute); independent from 2021 |
| Known for | AI governance research, policy-adjacent conceptual work on compute, corporate governance, safety cases, and international coordination |
Overview
GovAI was founded within the Future of Humanity Institute (FHI) at Oxford in 2018, led by Allan Dafoe. It spun out as an independent nonprofit in 2021 and has operated separately since, including through the April 2024 closure of FHI itself. Its alumni network has populated other AI governance institutions, including US AISI, UK AISI, Center for Security and Emerging Technology (CSET), METR, and frontier labs.
GovAI describes its mission as building the field of AI governance: developing the conceptual foundations, empirical research base, and trained talent to allow governments and institutions to handle transformative AI. Its work bridges technical research, policy design, and institutional analysis. Relative to peer institutions, it is more academic than Center for Security and Emerging Technology (CSET), more governance-focused than Apollo Research or METR, and more empirical than Machine Intelligence Research Institute (MIRI), and it is non-lab-aligned despite having alumni across all major labs. This positioning has made it a frequent reference for both governments and industry.
Research
GovAI's research spans several themes. Its work on frontier model safety cases develops structured safety arguments for frontier AI deployment. Its compute governance work, associated with the period when Lennart Heim led that area, argued that compute is the most tractable AI governance lever. The corporate AI governance theme examines internal governance structures at frontier labs. Its international coordination work covers institutional designs for multilateral AI governance and verification regimes, and its pre-deployment testing and evaluation work links technical evaluation to regulatory regimes.
The flagship publication is Safety Cases for Frontier AI (Buhl, Sett, Koessler, Schuett, Anderljung, 2024), which proposes adapting safety cases from the aviation and nuclear industries to frontier AI. GovAI research has also exerted upstream conceptual influence on Open Problems in Technical AI Governance, with several of its listed open problems originating in GovAI work. Foundational compute-governance work by Lennart Heim, then at GovAI and now at RAND Corporation, underlies the Compute Governance concept.
People
- Allan Dafoe — founder; now leads governance work at Google DeepMind
- Ben Garfinkel — former director
- Markus Anderljung — head of policy; co-author of Safety Cases for Frontier AI
- Lennart Heim — former GovAI compute-governance lead; now at RAND Corporation
- Miles Brundage — adjacent alumnus; former OpenAI policy head
GovAI alumni also populate US AISI, UK AISI, Center for Security and Emerging Technology (CSET), METR, and frontier labs.
Funding and governance
GovAI is funded by foundations in the broader AI-safety philanthropic ecosystem, notably Open Philanthropy, Longview Philanthropy, and the Survival and Flourishing Fund. It retains some research partnerships with Oxford and other UK universities.
Relationships
- related: Safety Cases for Frontier AI — flagship GovAI publication
- related: Open Problems in Technical AI Governance — cites GovAI-adjacent work
- related: Lennart Heim — GovAI alumnus; compute-governance lead
- related: Center for Security and Emerging Technology (CSET), RAND Corporation, Center for Strategic and International Studies (CSIS) — adjacent policy-research institutions; GovAI is the UK/Oxford node
- related: US AI Safety Institute (NIST AISI), UK AI Safety Institute (AI Security Institute) — alumni destinations
- related: Open Philanthropy — funder
- supports: Compute Governance, AI Safety Cases and Frameworks, AI Benchmarks and Evaluation