P062 Integrity and attack surface of the verification apparatus
Can the verifier-side infrastructure itself (recomputation servers, taps, sensors) be trusted and resist a state attacker, and how much attack surface does it add?
Whether verifier-side infrastructure (recomputation servers, taps, sensors, the regime as a whole) can itself be trusted and resist state attackers.
Status
Analyzed Design only, from a self-deprecated working draft (June 2026, since rewritten; treat as dated): a verifier-side auditing environment built from three independently sourced trusted computing bases whose outputs are compared before leaving an air-gap, plus an instructor-executor split with formally verified logic for the simple parts. Nothing built; no estimate of added attack surface. Recomputation-server hardening has no work (AI 2040: 'needs significant research'; Amodo: 'has not been explored'). Comer et al. 2026 (RAND) design a secure inference facility against nation-state attackers, but prover-side, so a cross-reference.
At least three distinct TCBs (different supply chains, hardware and software. One trusted by the prover, the verifier and a third-party) make up three small clusters, all inside the prover’s datacenter and inside their own air-gap.
Posed by
- Harack et al. (2025), Verification for International AI Governance (Oxford AIGI):
must be relatively continuous—therefore opening up further avenues for attack
(§2.5.1, p.58, L3161–3162) (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); Scher & Thiergart (2024), Mechanisms to Verify International Agreements About AI Development (MIRI); Petrie & Aarne (2025), Technical Options for Flexible Hardware-Enabled Guarantees (flexHEG Part 2); Shavit (2023), What Does It Take to Catch a Chinchilla? Verifying Rules on Large-Scale Neural Network Training via Compute Monitoring; Ilhan, Withers, Gietz & Harack (2026), Verifiable Semiconductor Manufacturing (Oxford AIGI); Cankaya (2026), A System Overview for Near-Term, Low-Trust AI Compute Verification (v0.2); Cankaya (2026), All Hands on Deck to Build the Datacenter Lie Detector; AI 2040 (2026), Get Involved in Verification; AI Security Forum, Requests for Discussion; Amodo (2026), AI 2040 Plan A — How You Can Help; 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 |
|---|---|---|
| P062.1 | Has an open-source tap or sensor design been jointly vetted by two adversarial parties, including physical scanning of production units, for hidden espionage or sabotage functions? | open |
| P062.2 | Does current GPU confidential computing / attestation (e.g. SPDM sessions) use post-quantum key exchange, closing harvest-now-decrypt-later? | partly answered |
| P062.3 | Can a recomputation server inside the prover's facility be hardened so the verifier detects integrity attacks (e.g. GPU firmware substitution) by a prover with physical access? | open |
| P062.4 | What minimum compute (H100-hours/day) does sampled recomputation need to reach a stated detection probability for a stated falsification rate? | open |
Sub-questions not yet stated precisely enough to check
| ID | Sub-question |
|---|---|
| P062.5 | How much exploitable attack surface do verification infrastructure and its continuous data flows add, and can it be bounded? |
| P062.6 | How can verification tools, including parts that cannot be shown to the counterparty, themselves be verified ('who watches the watchers')? |
| P062.7 | Can the verification regime, at full (e.g. 2034) scale, withstand nation-state hardware, software and supply-chain attacks? |
| P062.8 | What does it cost (time, money, IP disclosure) to disclose, authenticate, test and patch designs across the full stack so that stacks become mutually verifiable? |
| P062.9 | Is formally verified logic achievable for the trusted components needed for unilateral security against software attacks? |
| P062.10 | What residual risk counts as 'secure enough' given high-impact, low-probability attacks such as a single compromised chip? |
| P062.11 | How should assurance be tiered by chip criticality in a defence-in-depth design? |
| P062.12 | Can mechanisms built for today's threats withstand highly capable AI (new chip-production routes bypassing supply-chain monitoring, broken cryptography)? |
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 P062.2: Does current GPU confidential computing / attestation (e.g. SPDM sessions) use post-quantum key exchange, closing harvest-now-decrypt-later?
For P062.9: Is formally verified logic achievable for the trusted components needed for unilateral security against software attacks?
* 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