P071 Detecting fabricated logs
Which log-fabrication attacks are hardest to detect, and how much cross-device correlation is needed to catch them?
Detecting fabricated or incomplete activity logs from an operator with root and physical access: the hardest attacks, cross-device correlation, tamper evidence, logging standards and binding data to a chip.
Status
Posed Idea stage only: a proposed red-team/blue-team exercise to test whether cross-device log analysis catches fabricated logs, not yet run; no design for detecting fabrication.
Posed by
- AI Security Forum, Requests for Discussion:
What log fabrication attacks are hardest to detect?
(RFD 011 Nonconfidential Verification (Petrie), Open Questions; line 435) - AI Security Forum, Requests for Discussion:
How much cross-device correlation is needed to catch fabrication?
(RFD 011 Nonconfidential Verification (Petrie), Open Questions; line 436)
Also raised in: Scher (2026), Verifying Restrictions on Frontier AI Research (MIRI); Cankaya (2026), Compute Verification FAQs; Wasil, Reed, Miller & Barnett (2024), Verification Methods for International AI Agreements; Proofworks, Living Verification Documents.
Sub-questions
| ID | Sub-question | What the sources report |
|---|---|---|
| P071.1 | In red-team trials, what fraction of fabricated log sets are caught, which attacks are hardest, and how many correlated devices are needed? | open |
| P071.2 | Can chip activity logs be made tamper-evident against an operator with root and physical access? | implied by a source, not stated as open |
| P071.3 | Can storage-fabric taps or a signed workload ledger detect a checkpoint write or workload omitted from the commitment log? | implied by a source, not stated as open |
| P071.4 | Has a standards body published an AI-cluster verification logging protocol? | open |
Sub-questions not yet stated precisely enough to check
| ID | Sub-question |
|---|---|
| P071.5 | How can input data be bound to the specific chip/cluster that produced it? |
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 P071.2: Can chip activity logs be made tamper-evident against an operator with root and physical access?
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