P039 Power accounting at declared facilities
What error does power accounting give for the count of operating AI chips or the size of a run at a declared facility, under adversarial choice?
Power accounting of chips and runs at declared facilities.
Terms: β (beta): the largest hidden computation a power trace cannot rule out, as a fraction of the declared machine capacity; β (beta) is its worst case over declared utilisation (how busy the declared work keeps the chip). The paper's headline 1.16 is that worst case on the A100s measured; 0.41 is what a demonstrated attack actually hid.
Status
Started Measured on one NVIDIA A100 (energy read from the on-die counter, not a feed meter): a power trace alone cannot rule out between 0.41 and 1.16 machines of undeclared compute per declared machine, against an adversary who chooses numeric format, operand values and idle state; same-SKU transfer checked on three more A100s. One device only, so the facility-level, mixed-fleet and run-size clauses of the headline have no measured bound; time-resolved signals (training-step periodicity) are out of the paper's scope. Lower stages: Baker et al.'s PDU metering and three compute-accounting options (RAND 2025, App. A.6) are a design only; the synchronous-swing and shutdown-verification clauses are stated judgements with no work. A companion simulation study by the same group (October 2026) tracks the training-step rhythm in facility power instead of total energy; it detects synchronous training far more reliably but does not say how much compute is running.
"Measurements on NVIDIA A100 GPUs constrain β = 1.16 in the worst case, while adversarial matched-energy strategies are shown to hide at least β = 0.41 of compute."
Posed by
- Scher & Thiergart (2024), Mechanisms to Verify International Agreements About AI Development (MIRI):
This is likely spoofable with small performance penalties, e.g., by slowly adding more servers to a workload or by using various distributed training techniques.
(Building Blocks › Synchronous changes, p. 90, L4323–4325) (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); Baker, Kulp, Marks, Brundage & Heim (2025), Verifying International Agreements on AI: Six Layers of Verification for Rules on Large-Scale AI Development and Deployment (RAND WR-A4077-1); Wasil, Reed, Miller & Barnett (2024), Verification Methods for International AI Agreements.
Sub-questions
| ID | Sub-question | What the sources report |
|---|---|---|
| P039.1 | In a mixed-fleet data centre (old/new, AI/non-AI chips), what error does power accounting give for the count of operating AI chips? | open |
| P039.2 | What is the smallest training run, or the smallest excess over declared energy, that facility-level metering detects at a stated false-alarm rate? | open |
| P039.3 | Has PDU-level metering been used to count active accelerator-hours in a pilot, with tamper evidence? | a source suggests an answer (not checked) |
| P039.4 | Could a provable cap on a state's AI-attributable power (MW) bound its inference capacity within a stated factor? | open |
Sub-questions not yet stated precisely enough to check
| ID | Sub-question |
|---|---|
| P039.5 | Is a no-colocation rule (no power-hungry non-AI workloads or military sites in the same facility) needed for power- and inspection-based checks to work? |
| P039.6 | Are tens-of-MW synchronous swings (training start, checkpoints) a usable signal, or spoofable at small cost by ramping or distributed training? |
| P039.7 | Can emergency shutdown or unplugging be verified from facility power, backup-generation knowledge and thermal imaging? |
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 P039.2: What is the smallest training run, or the smallest excess over declared energy, that facility-level metering detects at a stated false-alarm rate?
For P039.6: Are tens-of-MW synchronous swings (training start, checkpoints) a usable signal, or spoofable at small cost by ramping or distributed training?
* 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