P040 Grid-side detection of small hidden loads
What is the smallest undeclared AI load detectable from grid or utility interval data at a fixed false-alarm rate?
Detecting small hidden AI loads from grid and utility data.
Terms: feeder: a distribution power line serving an area. phase unbalance: unevenness across the three phases of AC power, a tell-tale of some loads. CUSUM: a statistical test that flags a sustained shift in a signal.
Status
Analyzed A desk reference-class analysis of utility electricity records by facility size, using Pilz & Heim's data-centre counts and EIA survey counts: a 10 MW continuous load is visible at the utility level but blends with crypto-mining, continuous-process industrial and campus loads; at 1 MW and 100 kW the paper places the load among flat-load commercial customers "of which the United States has millions", a figure it takes from a count of commercial buildings by floor area (EIA CBECS 2018, Table B6), not by electric load or load factor. No detector built, no interval data analysed and no false-alarm rate, so this survey's false-alarm goal of 0.01 per facility-year is not addressed; assumes utility cooperation; off-grid generation defeats it.
"A 100 kW continuous load fits within the existing population of mid-size commercial customers with refrigeration, telecom, or 24/7 operations, of which the United States has millions (U.S. Energy Information Administration 2022b). There is no externally distinguishing power signature at this scale."
Posed by
- Scher & Thiergart (2024), Mechanisms to Verify International Agreements About AI Development (MIRI):
This is likely not the case for the distributed training threat model—a data center with 5,000 H100 GPUs has a power requirement of ~6 MW
(App. Track AI Chips, Not Data Centers, p. 141, L6576–6577) (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); Canary Institute research; Koopmanschap & Barten (2026), How to Catch a GPU: A Taxonomy of Verification and Enforcement Mechanisms for International AI Agreements; Wasil, Reed, Miller & Barnett (2024), Verification Methods for International AI Agreements; Proofworks, Living Verification Documents.
Sub-questions
| ID | Sub-question | What the sources report |
|---|---|---|
| P040.1 | What is the smallest undeclared AI load (MW) that grid or utility data have detected in a blind or field test? | partly answered |
| P040.2 | At what constant kW does a single premises' interval record become a statistical outlier among similar customers (a flat load hides in feeder aggregates)? | open |
| P040.3 | At a stated distribution-bus voltage, what fraction of hours is a training island of 16 H100-equivalents detected at a fixed false-alarm rate? | partly answered |
| P040.4 | Are the Elsayed jobs detected in every hour on the phase-unbalance signal at the LBNL bus? | a source suggests an answer (not checked) |
| P040.5 | Does the re-pulled LBNL record reproduce the 3.7-3.9x motor-start ratio (a figure from the internal record)? | partly answered |
| P040.6 | What detection rate does the pulse detector reach for a job split over 2 vs 3 phases? | open |
| P040.7 | Logged per-phase at a stated rate on PDU mains during multi-GPU training, what is the pulse depth on each phase? | open |
| P040.8 | Does a k-consecutive-hours rule remove the stove confounder on UK feeders, and at what cost in detection? | open |
| P040.9 | What are measured (not bounded) false-alarm rates of the feeder detectors on held-out data, per aggregation level? | partly answered |
| P040.10 | On real interval data, what is the CUSUM detection delay for a persistent 12 kW load at a fixed false-alarm rate? | partly answered |
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 P040.1: What is the smallest undeclared AI load (MW) that grid or utility data have detected in a blind or field test?
For P040.7: Logged per-phase at a stated rate on PDU mains during multi-GPU training, what is the pulse depth on each phase?
For P040.9: What are measured (not bounded) false-alarm rates of the feeder detectors on held-out data, per aggregation level?
For P040.10: On real interval data, what is the CUSUM detection delay for a persistent 12 kW load at a fixed false-alarm rate?
* 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