Positive and negative claims
A positive claim asserts that something is present or happened; a negative claim, that an activity or resource is absent; a mixed claim, both.
A positive claim asserts that something is present or happened, and a negative claim asserts that an activity or resource is absent 1; a mixed claim bundles both.
Examples are the declared model being the one served (positive), there being no undeclared compute (negative) and compute running inference and not training (mixed). The Oxford Martin report observes that demonstrating the existence of an object or process is often straightforward compared with demonstrating its non-existence 1. RAND's framework mirrors the split: one subgoal verifies that declared uses of compute are accurate, and another verifies that there are no undeclared uses and no undeclared clusters 2. Proposed designs support negative claims indirectly:
- Leaving no spare capacity. Filling or wiping memory removes residual capacity for hidden workloads, as in memory wiping and proofs of secure erasure 3.
- Limiting communication. Capping the bandwidth between pods at what inference tokens need, below what training gradients need, is meant to keep pods from joining a larger training run, as in bandwidth limits 4.
- Sampling. Randomly inspecting accelerators makes it likely that at least one accelerator used in a violating run is found 5.
All three act on declared hardware; Scher and Thiergart judge that detecting data centres that were never declared may be difficult 4.
Related
- Undeclared compute
- Sampling and assurance
- The declared model is the one being served
- This compute runs inference, not training
Used in
- R1Remote detection of data centres
- Compute stock is at most a declared amount
- Declared hardware is idle or shut down
- There is no undeclared relevant compute
Sources
- BB. Harack et al. (2025). Verification for International AI Governance. Oxford Martin AI Governance Initiative. Source recordSupports: demonstrating the existence of an object or process is often straightforward compared with demonstrating its non-existence · p. 31
- BM. Baker et al. (2025). Verifying International Agreements on AI: Six Layers of Verification for Rules on Large-Scale AI Development and Deployment. RAND Corporation. Source recordSupports: split between verifying that declared uses are accurate (Subgoal 1) and verifying no undeclared uses or clusters (Subgoal 2) · §3.2, Figure 4
- BN. Cankaya (2026). A System Overview for Near-Term, Low-Trust AI Compute Verification. Machine Intelligence Research Institute. Source recordSupports: memory wiping with incompressible noise to leave no residual capacity for hidden workloads · system architecture (memory wiping)
- BA. Scher & L. Thiergart (2025). Mechanisms to Verify International Agreements About AI Development. arXiv. Source recordSupports: bandwidth limits target the gap between inference tokens and training gradients between pods; covert data centres may be difficult to detect · Interconnect bandwidth limits; Verifying the location of AI compute
- BY. Shavit (2023). What does it take to catch a Chinchilla? Verifying Rules on Large-Scale Neural Network Training via Compute Monitoring. arXiv. Source recordSupports: random chip sampling detects at least one chip from a violating run with a chosen probability · §3.2, Equation 1