Organization · Academic
Oxford Martin AI Governance Initiative
A research programme at the University of Oxford that studies AI governance from technical and policy angles and hosts the Hardware AI Governance Lab.
aigi.ox.ac.uk · also called AIGI; Oxford Martin AIGI
The Oxford Martin AI Governance Initiative (AIGI) is a research programme housed in the Oxford Martin School at the University of Oxford 1. It describes itself as one of the few centres in the world that study the governance of AI from both technical and policy perspectives 1. It is co-led by Robert Trager, a social scientist who specialises in international relations and frontier AI regulation, and Maike Osborne, a machine-learning specialist 1.
Its verification work includes:
- A 2025 AIGI report examines how states could verify compliance with international AI agreements, framed around a prover and a verifier 2. It observes that demonstrating that something exists is often easier than demonstrating that it does not 2. It includes appendices on model fingerprint attestation, and describes networking hardware that makes cryptographic commitments about all traffic 2.
- A 2024 policy paper argues that compute providers could be verifiers of their customers' activities 3. It describes how information that providers already hold, without confidential data, can indicate a workload's type, such as large-scale training or inference, and the compute it consumed 3; see This compute runs inference, not training.
- AIGI hosts the Hardware AI Governance Lab, which studies how computer hardware architecture can support AI governance, oversight and international coordination 4.
- SASH reports launching, with AIGI and a team of researchers, an international collaboration to build tools that make AI verifiable across borders, and names AIGI as an early partner 5. The collaboration's first project is a confidential network logger 5; see SASH confidential network logger.
Publications
Sources this organization authored or published.
- BAbout: Oxford Martin AIGI. Oxford Martin AI Governance Initiative. RecordCited by Oxford Martin AI Governance Initiative
- BA. Ilhan et al. (2026). Verifiable Semiconductor Manufacturing. Oxford Martin AI Governance Initiative. RecordCited by Hardware AI Governance Lab
- BB. Harack et al. (2025). Verification for International AI Governance. Oxford Martin AI Governance Initiative. RecordCited by The declared model is the one being served; There is no undeclared relevant compute; Cryptographic commitment; Positive and negative claims; Prover; Threat model; Verifier; Oxford Martin AI Governance Initiative
- BG. Sastry et al. (2024). Computing Power and the Governance of Artificial Intelligence. arXiv. RecordCited by Communication between compute groups is bounded; Compute stock is at most a declared amount; This compute runs inference, not training; There is no undeclared relevant compute; A training run stayed within declared limits; Compartmentalization; FLOP accounting; Inference and training workloads; Interconnect bandwidth; Undeclared compute
- BL. Heim et al. (2024). Governing Through the Cloud: The Intermediary Role of Compute Providers in AI Regulation. Oxford Martin AI Governance Initiative. RecordCited by Oxford Martin AI Governance Initiative
- BR. Trager et al. (2023). International Governance of Civilian AI: A Jurisdictional Certification Approach. Centre for the Governance of AI. Record
Mentioned in
Sources
- BAbout: Oxford Martin AIGI. Oxford Martin AI Governance Initiative. Source recordSupports: housed in the Oxford Martin School; technical and policy focus; co-led by Robert Trager and Maike Osborne
- BB. Harack et al. (2025). Verification for International AI Governance. Oxford Martin AI Governance Initiative. Source recordSupports: Verification for International AI Governance: prover and verifier, presence versus absence, model fingerprint attestation, traffic commitments
- BL. Heim et al. (2024). Governing Through the Cloud: The Intermediary Role of Compute Providers in AI Regulation. Oxford Martin AI Governance Initiative. Source recordSupports: compute providers as verifiers; workload type and compute from non-confidential information
- BHardware AI Governance Lab. Oxford Martin AI Governance Initiative. Source recordSupports: hosts the Hardware AI Governance Lab; the lab's research focus
- CSingapore AI Safety Hub (SASH) (2026). Internationalising AI Verification. SASH blog. Source recordSupports: co-launched SASH's international verification collaboration; named as an early partner