# AI verification proposal A proposal built with the Proposal Explorer of the AI Verification Tech Map (https://trustbutveri.fyi/), from its records of 2026-10-09. Interactive version: https://trustbutveri.fyi/explorer/?mechanisms=M-0001,M-0007&implementations=M-0001:I-0002 How to read it: a claim is something one party wants to verify about another's AI hardware or software. A mechanism is a general technique for verifying claims; it is "aimed at" a claim when that is its direct purpose, and "supporting" when it contributes without being aimed at it. A claim is addressed when a mechanism in the proposal is aimed at it and is not excluded by the filters; addressed does not mean verified, so check that mechanism's development status, security evidence and findings. Definitions: https://trustbutveri.fyi/about/methodology/ (roles, properties and findings) and https://trustbutveri.fyi/about/readiness/ (development status). ## Filters Filters apply to mechanisms only and describe the setting the proposal is for. None set. Every mechanism on the map was available. ## Overview One row per mechanism, read from its record. Open failures: critical / significant / minor. The last three columns are the editors' reading of what the verifier sees. Findings are grouped as known failures, scope limitations and open questions. Only known failures count as failures. Counts are an inventory of published findings, not a risk score. | Mechanism | Development | Security evidence | Prover | Attack testing | Hardware | Open failures | Weights | Inputs and outputs | Training data | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | Sampled inference recomputation / DiFR (Divergence From Reference) | Research demonstration | Published security analysis | Adversarial | Analysis | None | 0 / 1 / 1 | unspecified | unspecified | unspecified | | Proofs of useful work for capacity accounting | Proposed | Published security analysis | Adversarial | Analysis | None | 0 / 1 / 0 | depends | depends | not involved | ## Claims No claims chosen. ## Mechanisms ### Sampled inference recomputation DiFR checks that an inference provider ran its declared model by comparing output tokens or activations with a trusted re-run using the same random seed. ([Sampled inference recomputation](https://trustbutveri.fyi/mechanisms/sampled-inference-recomputation/)) - Assessment: selected implementation [DiFR (Divergence From Reference)](https://trustbutveri.fyi/implementations/difr/). - Development: Research demonstration (legacy code R2), assessed for checking that outputs match the declared model, precision and sampling settings. - Security evidence: Published security analysis. Independent evaluation: unassessed. Formal proof: unassessed. Deployment assurance: unassessed. - Claims in this proposal: none of them. - Threat model: adversarial prover. Hardware: none. Prover cooperation: required. Attack testing: analysis. Category: Cryptographic & computational. - What the verifier sees: model weights unspecified; inputs and outputs unspecified; training data unspecified. This Explorer has no asset-specific exposure assessment for this implementation. Check its source and deployment assumptions. ### Proofs of useful work for capacity accounting Cryptographic evidence that a given amount of matrix-multiplication work was completed, proposed as one input to bounding how much spare capacity declared hardware has. ([Proofs of useful work for capacity accounting](https://trustbutveri.fyi/mechanisms/proofs-of-useful-work/)) - Assessment: mechanism family. - Development: Proposed (legacy code R1), assessed for bounding the spare capacity of declared hardware that could run training. - Security evidence: Published security analysis. Independent evaluation: unassessed. Formal proof: unassessed. Deployment assurance: unassessed. - Claims in this proposal: none of them. - Threat model: adversarial prover. Hardware: none. Prover cooperation: required. Attack testing: analysis. Category: Cryptographic & computational. - What the verifier sees: model weights depends; inputs and outputs depends; training data not involved. Checking a sampled tile of a matrix multiplication reveals that tile, which may hold model or input data; the authors suggest a zero-knowledge proof when the matrices must stay private. ## Properties **Built for an adversarial prover** - Sampled inference recomputation - Proofs of useful work for capacity accounting **No new hardware needed** - Sampled inference recomputation - Proofs of useful work for capacity accounting ## Attack testing Attack testing records published testing for this use. It does not by itself show independent review, a formal proof or that a deployed system is secure. **Testing history** - Sampled inference recomputation / DiFR (Divergence From Reference): Analysis - Proofs of useful work for capacity accounting: Analysis ## Limits **Open significant failures** - Statistical tolerance leaves a covert channel (known failure, demonstrated attack, in DiFR (Divergence From Reference); https://trustbutveri.fyi/implementations/difr/evidence/flaws/1/) [2][7][9]. Statistical schemes can put an upper bound on an adversary's covert bandwidth, but cannot close it. In the companion exfiltration study, the detector cut exfiltratable information to under 0.5% under benign prompt traffic. It did not cut it to zero. Its authors called the channel's size under adversarial prompts an open empirical question. An independent study attacked that detector, which uses the same Gumbel-margin statistic. An adversary who controls the prompts roughly doubled the bits leaked per token. Across six models, this cut the slowdown from 146–254 times under benign prompts to 60–118 times. The attack widens the exfiltration bound. It does not target DiFR's check that outputs match the declared configuration. - Known shortcuts let a miner claim somewhat more work than it did (known failure, theoretical argument, in Proofs of useful work for capacity accounting; https://trustbutveri.fyi/mechanisms/proofs-of-useful-work/evidence/flaws/3/) [15]. Pearl's specification lists known mining speedups: crafted inputs, precision shortcuts, seed grinding, work reuse, and faster kernels or hardware. A policy check caps the summands a miner may skip at one-sixteenth of those in a tile. For capacity bounding, any gap between work proven and work possible leaves spare capacity. **Family finding context** - Context for DiFR (Divergence From Reference). Findings from the mechanism family appear here as context. They apply to an implementation only when its own record lists them, under the conditions stated there. Tolerance for numerical noise leaves a covert channel (known failure, demonstrated attack, in Sampled inference recomputation; https://trustbutveri.fyi/mechanisms/sampled-inference-recomputation/evidence/flaws/1/) [2][7][9]. Schemes that accept approximate matches can put an upper bound on an adversary's covert bandwidth, but they cannot close the channel. The weight-exfiltration detector cut exfiltratable information to under 0.5%, not to zero, on a 30-billion-parameter mixture-of-experts model under benign prompt traffic. Its authors called the channel's size under adversarial prompts an open empirical question. An independent study showed that an adversary who controls the prompts roughly doubles the bits leaked per token. Across six models, that cut the slowdown from 146–254 times under benign prompts to 60–118 times. The attack widens the exfiltration bound. It does not target the check that outputs match the declared model. Related mechanism: Deterministic and bit-exact inference (R3, not in the proposal). Bit-exact inference would remove the numerical tolerance if exact replay can be deployed with the required weights and configuration. - Context for DiFR (Divergence From Reference). Findings from the mechanism family appear here as context. They apply to an implementation only when its own record lists them, under the conditions stated there. Only recorded traffic is checked (scope limitation, theoretical argument, in Sampled inference recomputation; https://trustbutveri.fyi/mechanisms/sampled-inference-recomputation/evidence/flaws/2/) [2][10]. Recomputation checks that recorded, declared workloads are correct. It cannot show that the record is complete. The published schemes do not cover hidden workloads run on the same compute, or substituted work. Rinberg et al. say their exfiltration-detection scheme cannot stand alone. Related mechanism: Network taps and certifiers (R1, not in the proposal). Taps copy and hash all traffic on the monitored links, which bears on whether the traffic record is complete. They do not show what else ran on the same chips. - Context for DiFR (Divergence From Reference). Findings from the mechanism family appear here as context. They apply to an implementation only when its own record lists them, under the conditions stated there. Some inference optimizations are not covered (known failure, theoretical argument, in Sampled inference recomputation; https://trustbutveri.fyi/mechanisms/sampled-inference-recomputation/evidence/flaws/3/) [1][11]. TOPLOC's authors state that it cannot detect speculative decoding in which a cheaper model does the decoding. They did not test whether it distinguishes types of key-value (KV) cache compression. DiFR was evaluated only on sampling from a single model. Its authors sketch an extension to one speculative-decoding algorithm but do not test it. - Context for DiFR (Divergence From Reference). Findings from the mechanism family appear here as context. They apply to an implementation only when its own record lists them, under the conditions stated there. Mixed hardware widens the honest baseline (known failure, open question, in Sampled inference recomputation; https://trustbutveri.fyi/mechanisms/sampled-inference-recomputation/evidence/flaws/4/) [1]. When honest reference runs span different GPU types, the spread of benign scores grows. In DiFR's tests on Qwen3-30B-A3B, pooling A100 and H200 runs left Token-DiFR unable to separate the two smallest tested changes, a temperature of 1.1 instead of 1.0 and a simulated top-2 sampling bug, at the target false-positive rate, while cross-entropy separated them. Matched provider and verifier environments, or pooling that weights rare large deviations, restored detection. **Open minor failures** - Mixed hardware widens the honest baseline (known failure, open question, in DiFR (Divergence From Reference); https://trustbutveri.fyi/implementations/difr/evidence/flaws/2/) [1]. For Qwen3-30B-A3B, pooling honest runs across A100 and H200 GPUs and parallelism setups broadened the honest score distribution. Token-DiFR then failed to separate the two smallest tested changes, a temperature of 1.1 instead of 1.0 and a simulated top-2 sampling bug, at the target false-positive rate, while cross-entropy did. The authors report that matched provider and verifier environments, or pooling that weights rare large deviations, restore detection. **Scope limitations** - Proves that work was done, not that no capacity remains (scope limitation, theoretical argument, in Proofs of useful work for capacity accounting; https://trustbutveri.fyi/mechanisms/proofs-of-useful-work/evidence/flaws/1/) [12]. Proof-of-work accounting bounds unmonitored compute only relative to an estimate of what the actor has. Attestable states that the verifier "needs a credible estimate of the compute available" to the actor, and that a proof "cannot discover a datacenter that was never declared". Related mechanism: Chip registries and manufacturing records (R1, not in the proposal). A registry of chips is one basis for the estimate of available compute that the flaw's source says the verifier needs. Related mechanism: Remote detection of data centres (R1, not in the proposal). Looks for data centres that were never declared, which a proof cannot discover. **Open questions** - Speculative decoding and multi-model sampling not evaluated (open question, open question, in DiFR (Divergence From Reference); https://trustbutveri.fyi/implementations/difr/evidence/flaws/3/) [1]. The algorithms and experiments cover sampling from a single LLM. Speculative decoding was not evaluated. The authors sketch an extension to one speculative-decoding algorithm, without experiments. They note that other variants would need modified verification and extra metadata. - Security rests on new hardness assumptions (open question, open question, in Proofs of useful work for capacity accounting; https://trustbutveri.fyi/mechanisms/proofs-of-useful-work/evidence/flaws/2/) [14][15]. Komargodski and Weinstein base security on hardness assumptions about batches of low-rank random linear equations, and list PoUW "from more standard or well-studied assumptions" as an open problem. Pearl's floating-point variant introduces a further "quantized-subspace hardness" assumption. **Not yet demonstrated** - Proofs of useful work for capacity accounting: Proposed (legacy code R1), assessed for bounding the spare capacity of declared hardware that could run training ## Possible additions Mechanisms on the map, not in the proposal, that the records connect to an unaddressed or partly addressed claim, an open failure or a dependency. Pointers, not recommendations: each brings its own readiness level and findings, and none is claimed to close a failure. None found. ## Dependencies **Blockers** - Sampled inference recomputation: The verifier needs the model weights, so outsiders cannot use the method to verify providers of closed-weights models. (privacy & leakage) [1] - Sampled inference recomputation: The verifier must know and match the provider's sampling procedure, and in one prototype a sampling mismatch in a newer vLLM version produced large spurious logit differences. (performance & compatibility) [1][4] - Sampled inference recomputation: No independent red-team of DiFR's consistency check has been published, Amodo rates recomputation red-teaming 'not started', and the one independent attack study targets an exfiltration detector built on the same statistic. (adversarial validation) [6][7] - Proofs of useful work for capacity accounting: Bounding spare capacity needs a credible estimate of the compute available to the actor, including third-party access. (capacity bounds) [12] - Proofs of useful work for capacity accounting: Proofs of work cannot find facilities that were never declared. (coverage & hidden compute) [12] - Proofs of useful work for capacity accounting: As of September 2026 no implementation, demonstration or independent evaluation of proofs of work for capacity bounding has been published. (adversarial validation) ## What the verifier sees - Model weights: shown by none; depends on the design for Proofs of useful work for capacity accounting; hidden by none; not involved in none; unspecified for Sampled inference recomputation. - Inputs and outputs: shown by none; depends on the design for Proofs of useful work for capacity accounting; hidden by none; not involved in none; unspecified for Sampled inference recomputation. - Training data: shown by none; depends on the design for none; hidden by none; not involved in Proofs of useful work for capacity accounting; unspecified for Sampled inference recomputation. ## Implementations - Sampled inference recomputation: [AI 2040 inference-only verification stack](https://trustbutveri.fyi/implementations/ai-2040-inference-only-verification-plan/) (R1, proposed architecture); [DiFR (Divergence From Reference)](https://trustbutveri.fyi/implementations/difr/) (R2, research prototype); [Low-trust AI compute verification system overview](https://trustbutveri.fyi/implementations/low-trust-compute-verification-system-overview/) (R1, proposed architecture); [SASH confidential network logger](https://trustbutveri.fyi/implementations/sash-confidential-network-logger/) (R1, research prototype); [TOPLOC](https://trustbutveri.fyi/implementations/toploc/) (R3, open-source project) - Proofs of useful work for capacity accounting: [Pearl proof-of-useful-work blockchain](https://trustbutveri.fyi/implementations/pearl-proof-of-useful-work/) (R3, open-source project) ## Sources 1. DiFR: Inference Verification Despite Nondeterminism, A. Karvonen et al. (2025). https://arxiv.org/abs/2511.20621 2. Verifying LLM Inference to Detect Model Weight Exfiltration, R. Rinberg et al. (2025). https://arxiv.org/abs/2511.02620 3. adamkarvonen/difr (GitHub repository), A. Karvonen (2025). https://github.com/adamkarvonen/difr 4. Scaling Recomputation Inference Verification, Amodo Design (2026). https://amododesign.com/notes/2026-09-02-scaling-recomputation-inference-verification/ 5. Amodo-Design/Inference-Recomputation-Prototype (GitHub repository), Amodo Design (2026). https://github.com/Amodo-Design/Inference-Recomputation-Prototype 6. AI 2040 Plan A — Verification SITREP, Amodo Design (2026). https://amododesign.com/ai-verification/plan-a-sitrep/ 7. Adversarial Entropy Inflation Against Gumbel-Based Inference Verification, N. Kezins (2026). https://arxiv.org/abs/2608.23375 8. An Inference Verification Prototype — Stage 1, Amodo Design (2026). https://amododesign.com/notes/2026-06-29-inference-verification-prototype/ 9. Bit-Exact AI Inference Verification Without Performance Tradeoffs, N. Cankaya (2026). https://arxiv.org/abs/2606.00279 10. Example Schemes for Verifying High-Stakes AI Agreements, Amodo Design (2026). https://amododesign.com/notes/2026-06-23-verification-algorithms/ 11. TOPLOC: A Locality Sensitive Hashing Scheme for Trustless Verifiable Inference, J. M. Ong et al. (2025). https://proceedings.mlr.press/v267/ong25a.html 12. Pacing AI Requires Proof, Attestable (2026). https://attestable.com/blog/pacing-ai-requires-proof 13. Mechanisms to Verify International Agreements About AI Development, A. Scher & L. Thiergart (2025). https://arxiv.org/abs/2506.15867 14. Proofs of Useful Work from Arbitrary Matrix Multiplication, I. Komargodski & O. Weinstein (2025). https://arxiv.org/abs/2504.09971 15. Pearl Floating Point Scheme Specification, Pearl Research Team (2026). https://pearlresearch.ai/Pearl_Whitepaper.pdf 16. pearl: Monorepo for the Pearl network, Pearl Research Labs (2026). https://github.com/pearl-research-labs/pearl