P025 Digital perimeters, egress accounting and inference-site architecture
Can a digital-perimeter and egress-accounting design show, to stated confidence, that every outbound message from an inference site has a log entry?
Digital perimeters, egress accounting and inference-site architecture.
Status
Started Hashing every frame on one 400GbE link with generator traffic on real hardware: line rate holds for frames of 1500 B and above on a 32-core CPU, fails for 64 B frames and on DPU offload - a prover choosing small frames defeats complete hashing. This is the capture layer only; reconciling hashed egress against an inference log to a stated confidence, red-team exfiltration tests, and the inference-site architecture (weights path over partitioned networks and data diodes, where 'we have not seen any work') are paper only or no work. The designs are Cankaya's north-south taps with all egress 'fully explainable' by observed activity and Rinberg's Merkle inference ledger. The same benchmark is the evidence for the traffic-hashing problem.
For worst-case 64B minimum frames, no algorithm on this hardware reaches full line rate … For realistic traffic profiles at 1500B and above, AES-128-GMAC and SipHash-1-3-128 are both capable of sustaining line-rate hashing
Posed by
- Harack et al. (2025), Verification for International AI Governance (Oxford AIGI):
credibly demonstrating that it has not been copied is extremely difficult
(§1.5.3 fn 100, p.34, L1897–1898) (source states the problem but not the quantity the question asks for)
Also raised in: Rinberg et al. (2025), Verifying LLM Inference to Detect Model Weight Exfiltration; Peigné-Lefebvre, Nguyen & Wang (2026), Zero Knowledge Verification for Frontier AI Training Is Possible, arXiv 2606.05433; AI 2040 (2026), Get Involved in Verification; Amodo (2026), AI 2040 Plan A — Verification SITREP; Dean, Verification Plan (AI 2040); Proofworks, Living Verification Documents.
Sub-questions
| ID | Sub-question | What the sources report |
|---|---|---|
| P025.1 | Has a digital-perimeter design been tested against a red team exfiltrating a known file, with a reported detection rate and bandwidth floor (i.e. can one show data was not copied)? | open |
| P025.2 | Can a hardware or enclave mechanism emit a verifiable proof that each outbound blob was encrypted under a declared public key? | open |
| P025.3 | How would off-chip encryption hardware identify which outbound data to encrypt, and what is the throughput cost of encrypting all pod egress at line rate? | open |
| P025.4 | Can an egress tap reconciled against a log (e.g. a Merkle inference ledger) show to stated confidence that every outbound message has a log entry? | open |
| P025.5 | How much external bandwidth does frontier-scale inference need per served replica? | a source suggests an answer (not checked) |
| P025.6 | Can weights move from storage to inference units over a partitioned, data-diode path off the frontend network, with verifier-checkable hashes? | open |
| P025.7 | Can a tap-side whitelist verify a >=100 GB weight/server-image upload on the frontend network at 400G? | open |
| P025.8 | What throughput loss results from serving a frontier MoE model in isolated units with backend scale-out networking removed? | implied by a source, not stated as open |
Sub-questions not yet stated precisely enough to check
| ID | Sub-question |
|---|---|
| P025.9 | How much about in-datacenter activity can full egress observation establish? |
| P025.10 | Has an ablation of the secure network gateway / Secure ReComputer (inference-only package)'s security properties and its upload limits been done? |
| P025.11 | How can training be verified across sites when control of the linking network is split across administrative boundaries? |
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 P025.2: Can a hardware or enclave mechanism emit a verifiable proof that each outbound blob was encrypted under a declared public key?
For P025.4: Can an egress tap reconciled against a log (e.g. a Merkle inference ledger) show to stated confidence that every outbound message has a log entry?
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