P095 Hidden concurrent compute on declared hardware
Does a hidden concurrent workload produce a measurable signal, and how much compute can be freed undetectably on declared hardware?
Hidden concurrent compute and memory on declared hardware.
Status
Started Hidden inference on parallel CUDA streams alongside re-executed LLM inference on identical hardware, detected by timing with numerical fingerprinting as a complement; the author's own experiments, unquantified (no fraction of declared compute, no detection rate). The second clause, how much compute can be freed undetectably, has only paper-only work: Reuter's expectation for his own proposed protocol (1-2%) and a verifiable-memory-exhaustion draft with placeholder results. Bounding hidden memory and model-swap time: nothing built.
Any significant hidden workload would contend for resources, causing detectable slowdowns. In my experiments, resource contention was evident in timing.
Posed by
- Amodo (2026), Example Schemes for Verifying High-Stakes AI Agreements:
There are attacks on the schemes above that involve running hidden workloads on the same compute, and substituting work.
(Appendix 1, Completeness, L149) (source states the problem but not the quantity the question asks for)
Also raised in: 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); Cankaya (2025), Catching Misreporting About ML Hardware Use by Turning Noise into Signal; AI Security Forum, Requests for Discussion; Dean, Verification Plan (AI 2040); Proofworks, Living Verification Documents.
Sub-questions
| ID | Sub-question | What the sources report |
|---|---|---|
| P095.1 | Does a hidden concurrent workload (parallel streams, interleaved ops, spare capacity) of a stated fraction of declared FLOP cause a measurable signal, and how much compute can be freed undetectably? | partly answered |
| P095.2 | Can a memory fill-and-return or challenge function provably require the full declared model in memory, bounding hidden memory and model-swap time (incl. tensor-parallel layouts)? | open |
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 P095.1: Does a hidden concurrent workload (parallel streams, interleaved ops, spare capacity) of a stated fraction of declared FLOP cause a measurable signal, and how much compute can be freed undetectably?
For P095.2: Can a memory fill-and-return or challenge function provably require the full declared model in memory, bounding hidden memory and model-swap time (incl. tensor-parallel layouts)?
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