P053 Losses, outages and deniability
What is the baseline rate of accelerators lost/unaccounted in audited fleets, and what rule separates genuine losses from strategically labelled outages?
Chip losses, outages and plausible deniability.
Status
Posed
Posed by
- Petrie & Aarne (2025), Technical Options for Flexible Hardware-Enabled Guarantees (flexHEG Part 2):
any physical tampering would need to be either difficult to detect or limited to a small number of devices that could plausibly be reported as "lost" or otherwise unavailable for inspection.
(App B, Domestic Oversight, p55, L2535–2537) (source states the problem but not the quantity the question asks for)
Also raised in: Harack et al. (2025), Verification for International AI Governance (Oxford AIGI); Aarne & Petrie (2025), International Security Applications of Flexible Hardware-Enabled Guarantees (flexHEG Part 3); Shavit (2023), What Does It Take to Catch a Chinchilla? Verifying Rules on Large-Scale Neural Network Training via Compute Monitoring; Canary Institute research.
Sub-questions
| ID | Sub-question | What the sources report |
|---|---|---|
| P053.1 | What is the measured baseline annual rate of AI accelerators lost or unaccounted for in large audited fleets? | open |
| P053.2 | Can licences be verifiably denied to chips reported missing? | implied by a source, not stated as open |
| P053.3 | Can remote firmware-version attestation be defeated by 'lost in transit' claims, and what rule closes it? | open |
| P053.4 | Under each heartbeat downtime rule, what escape probability does an operator that strategically labels outages achieve? | open |
Sub-questions not yet stated precisely enough to check
| ID | Sub-question |
|---|---|
| P053.5 | When should a claimed large chip loss trigger investigation, and how is it run? |
| P053.6 | Can declaration flooding saturate a verification system's bandwidth? |
| P053.7 | Can states be relied on to report incidents to the international community? |
Possibly relevant work found by automated search
Found by AI agents searching for this problem's sub-questions, mostly by keyword and citation in the OpenAlex scholarly database and on the web; a second AI pass reviewed each one, and we spot-checked some by hand. They seemed reasonably related, as far as one can tell from outside each paper's field. Some come from neighbouring fields, such as arms control and financial audit, and bear on the problem by analogy. Not reviewed against the status above; a paper listed here does not change that status. 928 of 1052 quotes checked word for word against the paper's text.
For P053.1: What is the measured baseline annual rate of AI accelerators lost or unaccounted for in large audited fleets?
For P053.2: Can licences be verifiably denied to chips reported missing?
For P053.3: Can remote firmware-version attestation be defeated by 'lost in transit' claims, and what rule closes it?
For P053.4: Under each heartbeat downtime rule, what escape probability does an operator that strategically labels outages achieve?
For P053.5: When should a claimed large chip loss trigger investigation, and how is it run?
* Full text not reviewed; more likely to contain errors.
Data
- problems.csv — one row per problem
- problem_status.csv — status, scope and evidence per problem
- problem_sources.csv — the posed-by quote and other documents per problem
- leaves.csv — all sub-questions, including those without a checkable form
- leaf_sources.csv — every source row behind each sub-question