Source · Tier B · Technical reportSource · Verification for International AI Governance
Verification for International AI Governance
B. Harack, R. F. Trager, A. Reuel, D. Manheim, M. Brundage, O. Aarne, A. Scher, Y. Pan, J. Xiao, K. Loke, S. N. Adan, G. Bas, N. A. Caputo, J. C. Morse, J. Ahuja, I. Duan, J. Egan, B. Bucknall, B. Rosen, R. Araujo, V. Boulanin, R. Lall, F. Barez, S. Alvira, C. Katzke, A. Atamli, A. Awad. 2025. Oxford Martin AI Governance Initiative.
| Original | https://www.oxfordmartin.ox.ac.uk/publications/verification-for-international-ai-governance |
|---|---|
| Version | Sections 2.5.4.2 (printed pp. 69–71) and 4.5.2.3.6.c–d (pp. 122–124), and Appendices K and L.4–L.5 (pp. 157, 159–161, including footnote 437), read on 2026-10-21 for mobile enclosures, device-model mating, delayed operational logs, model-tenancy history and timed model fingerprint attestation. For the timed proposal, challenge construction is future work and about 50 ms is a proposed requirement. |
| Accessed | 2026-10-21 |
| Authored or published by | Oxford Martin AI Governance Initiative |
| Imported from | hodgkins-ai-verification-papers@c71e59ff0e8e |
| Note | Listed under "Research agendas and surveys" in the Hodgkins bibliography (CC BY 4.0). |
Cited by
- Hardware-attested weight bindingMechanism
- Tamper evidence for verifier devicesMechanism
- Timed challenge-response and memory-occupation challengesMechanism
- Model fingerprint attestationImplementation
- VRAM-residency challengeImplementation
- The declared model is the one being servedClaim
- There is no undeclared relevant computeClaim
- Declared safeguards were applied during inferenceClaim
- Cryptographic commitmentConcept
- Positive and negative claimsConcept
- ProverConcept
- Threat modelConcept
- VerifierConcept
- Oxford Martin AI Governance InitiativeOrganization