Source · Tier B · Technical reportSource · Verifying International Agreements on AI: Six Layers of Verification for Rules on Large-Scale AI Development and Deployment
Verifying International Agreements on AI: Six Layers of Verification for Rules on Large-Scale AI Development and Deployment
M. Baker, G. Kulp, O. Marks, M. Brundage, L. Heim. 2025. RAND Corporation.
| Original | https://www.rand.org/pubs/working_papers/WRA4077-1.html |
|---|---|
| Archived copy | https://web.archive.org/web/20260918125709/https://www.rand.org/pubs/working_papers/WRA4077-1.html |
| DOI | 10.7249/WRA4077-1 |
| arXiv | 2507.15916 |
| Accessed | 2026-10-20 |
| Authored or published by | RAND |
| Imported from | hodgkins-ai-verification-papers@c71e59ff0e8e |
| Note | Listed under "Research agendas and surveys" in the Hodgkins bibliography (CC BY 4.0). |
Cited by
- Deploy only evaluated modelsGoal
- Pause frontier AI developmentGoal
- Prevent catastrophic misuseGoal
- Chip registries and manufacturing recordsMechanism
- Remote detection of data centresMechanism
- Communication between compute groups is boundedClaim
- Compute stock is at most a declared amountClaim
- Declared hardware is idle or shut downClaim
- The declared model is the one being servedClaim
- This compute runs inference, not trainingClaim
- There is no undeclared relevant computeClaim
- Declared safeguards were applied during inferenceClaim
- A training run stayed within declared limitsClaim
- Model weights have not left the facilityClaim
- CompartmentalizationConcept
- FLOP accountingConcept
- Inference and training workloadsConcept
- Network tapConcept
- Positive and negative claimsConcept
- ProverConcept
- Undeclared computeConcept
- VerifierConcept
- RANDOrganization