# 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-0013&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 |
| Network taps and certifiers | Proposed | Published security analysis | Adversarial | Analysis | Retrofit device | 0 / 2 / 0 | depends | depends | depends |

## 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.

### Network taps and certifiers

Devices on a cluster's network links that copy and hash all traffic, so a verifier can later check sampled records against declared work. ([Network taps and certifiers](https://trustbutveri.fyi/mechanisms/network-taps-and-certifiers/))

- Assessment: mechanism family.
- Development: Proposed (legacy code R1), assessed for committing a complete record of cluster traffic, so declared inference can be checked.
- 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: retrofit device. Prover cooperation: required. Attack testing: analysis. Category: Off-chip devices & sensors.
- What the verifier sees: model weights depends; inputs and outputs depends; training data depends. Only hashes leave the site; records picked for a challenge are opened for replay at a verification facility.


## Properties

**Built for an adversarial prover**

- Sampled inference recomputation
- Network taps and certifiers

**No new hardware needed**

- Sampled inference recomputation

**Failures since mitigated**

- Verifier dictionary attacks on hashes (in Network taps and certifiers) [12]


## 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
- Network taps and certifiers: 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.
- Output nondeterminism leaves covert capacity (known failure, theoretical argument, in Network taps and certifiers; https://trustbutveri.fyi/mechanisms/network-taps-and-certifiers/evidence/flaws/1/) [7][12]. Hashing cannot remove information hidden in the outputs themselves. The Secure Gateway Device paper estimates that about 0.1 bit per token remains even with seed-synchronized replay checks. For a 200k-GPU inference cluster at full load (2,000 tokens per GPU per second), that is about 40 Mbit/s of covert egress, enough to move a 1 TB model in under three days. The paper names this the core remaining challenge and points to deterministic replay or active scrubbing of hardware-induced entropy. An independent study found that an adversary who chooses the prompts roughly doubles the bits leaked per token under Gumbel-based inference verification; see Bounding unexplained information in outputs.

  Related mechanism: Deterministic and bit-exact inference (R3, not in the proposal). Deterministic replay is one of the two remedies the flaw's source names.

  Related mechanism: Bounding unexplained information in outputs (R2, not in the proposal). Bounds the hidden information outputs can carry by measuring what the declared computation fails to predict.
- Residual side channels in simple passive setups (known failure, theoretical argument, in Network taps and certifiers; https://trustbutveri.fyi/mechanisms/network-taps-and-certifiers/evidence/flaws/4/) [16]. Amodo's analysis of its own tapped prototype lists unvalidated header fields, timing of permitted traffic and variation in response formatting as residual channels, and concludes that the passive tap must be replaced by an active one.

**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, 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**

- Some links cannot be passively tapped (scope limitation, open question, in Network taps and certifiers; https://trustbutveri.fyi/mechanisms/network-taps-and-certifiers/evidence/flaws/2/) [13][19]. Cankaya notes that copper-connected scale-up domains (for example NVL72 racks and TPU v7 cubes) are much harder to tap than fibre, and that optical budgets make passive taps impractical on 400GBASE-SR8 multimode links. Amodo found no taps advertised for 53 GBaud links as of May 2026.
- Encrypted fabrics hide plaintext from both parties (scope limitation, open question, in Network taps and certifiers; https://trustbutveri.fyi/mechanisms/network-taps-and-certifiers/evidence/flaws/3/) [13]. Cankaya notes that with TEE-protected sessions whose keys are ephemeral and managed inside the TEE, neither the operator nor the manufacturer can recover session keys after the session, so tapped traffic could not be opened for recomputation. For other encrypted fabrics, the operator can retain keys.
- Completeness rests on physical monitoring left out of scope (scope limitation, open question, in Network taps and certifiers; https://trustbutveri.fyi/mechanisms/network-taps-and-certifiers/evidence/flaws/5/) [12]. The Secure Gateway Device paper assumes the facility is physically monitored, and states that the whole architecture depends on the device being the only communication channel. It names radio emanation, power-line signalling and thermal channels as covert channels beyond that scope.

  Related mechanism: Side-channel suppression for isolated facilities (R1, not in the proposal). Addresses the radio, power-line and thermal channels that network-level designs leave out.

**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.

**Not yet demonstrated**

- Network taps and certifiers: Proposed (legacy code R1), assessed for committing a complete record of cluster traffic, so declared inference can be checked


## 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.

- **Deterministic and bit-exact inference** (Operational use (legacy code R3), assessed for reproducing open-model inference from receipts in Gensyn's information-market service)
  - Bears on the open significant failure "Output nondeterminism leaves covert capacity" in Network taps and certifiers. Deterministic replay is one of the two remedies the flaw's source names.
  - Network taps and certifiers waits on it: Nondeterministic inference leaves covert capacity in outputs that hashing cannot remove.
- **Bounding unexplained information in outputs** (Research demonstration (legacy code R2), assessed for bounding how much hidden information can leave in checked inference outputs)
  - Bears on the open significant failure "Output nondeterminism leaves covert capacity" in Network taps and certifiers. Bounds the hidden information outputs can carry by measuring what the declared computation fails to predict.
- **Tamper evidence for verifier devices** (Research demonstration (legacy code R2), assessed for detecting probing of proposed verifier hardware, using server and electronics prototypes as evidence)
  - Network taps and certifiers waits on it: Taps and gateway devices need tamper-evident housing and physical monitoring so that traffic cannot bypass them.
- **Side-channel suppression for isolated facilities** (Proposed (legacy code R1), assessed for bounding physical covert channels out of a verified enclosure)
  - Network taps and certifiers waits on it: Radio, power-line and thermal channels are not addressed by network-level designs.


## Dependencies

**Missing prerequisites**

- Deterministic and bit-exact inference (Operational use (legacy code R3), assessed for reproducing open-model inference from receipts in Gensyn's information-market service), needed by Network taps and certifiers
- Tamper evidence for verifier devices (Research demonstration (legacy code R2), assessed for detecting probing of proposed verifier hardware, using server and electronics prototypes as evidence), needed by Network taps and certifiers
- Side-channel suppression for isolated facilities (Proposed (legacy code R1), assessed for bounding physical covert channels out of a verified enclosure), needed by Network taps and certifiers

**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]
- Network taps and certifiers: No complete verification tap has been demonstrated at production frontend link rates, and on the tested CPU no hash algorithm reached line rate with minimum-size frames. (performance & compatibility) [19][20]
- Network taps and certifiers: Nondeterministic inference leaves covert capacity in outputs that hashing cannot remove. (evidence binding; waits on Deterministic and bit-exact inference) [12]
- Network taps and certifiers: Taps and gateway devices need tamper-evident housing and physical monitoring so that traffic cannot bypass them. (hardware trust; waits on Tamper evidence for verifier devices) [12][14]
- Network taps and certifiers: Radio, power-line and thermal channels are not addressed by network-level designs. (coverage & hidden compute; waits on Side-channel suppression for isolated facilities) [12]
- Network taps and certifiers: Red-teaming by specialists is called for but has not been reported. (adversarial validation) [12]


## What the verifier sees

- Model weights: shown by none; depends on the design for Network taps and certifiers; hidden by none; not involved in none; unspecified for Sampled inference recomputation.
- Inputs and outputs: shown by none; depends on the design for Network taps and certifiers; hidden by none; not involved in none; unspecified for Sampled inference recomputation.
- Training data: shown by none; depends on the design for Network taps and certifiers; hidden by none; not involved in none; 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)
- Network taps and certifiers: [AI 2040 inference-only verification stack](https://trustbutveri.fyi/implementations/ai-2040-inference-only-verification-plan/) (R1, proposed architecture); [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)

## 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. Fingerprinting All AI Cluster I/O Without Mutually Trusted Processors, N. Cankaya et al. (2026). https://arxiv.org/abs/2606.10724
13. The Fundamentals and Feasibility of Secure Network Taps for Verifying AI Datacenter Use, N. Cankaya (2026). https://nacicankaya.substack.com/p/research-note-the-fundamentals-and
14. A System Overview for Near-Term, Low-Trust AI Compute Verification, N. Cankaya (2026). https://intelligence.org/wp-content/uploads/2026/06/A-system-overview-for-near-term-low-trust-AI-compute-verification.pdf
15. Verification Plan, R. Dean (2026). https://ai-2040.com/supplements/verification-plan
16. Fitting a Network TAP to our Inference Verification Prototype, Amodo Design (2026). https://amododesign.com/notes/2026-09-15-network-tap-inference-verification/
17. inference-verification: Inference Verification Prototype, Singapore AI Safety Hub (SASH) (2026). https://github.com/sg-ai-safety-hub/inference-verification
18. Internationalising AI Verification, Singapore AI Safety Hub (SASH) (2026). https://www.aisafety.sg/research/internationalising-ai-verification
19. Network Tapping for AI Verification: A Technical Assessment, Amodo Design (2026). https://amododesign.com/notes/2026-05-03-network-tapping/
20. Network Traffic Hashing, Amodo Design (2026). https://amododesign.com/notes/2026-07-03-network-traffic-hashing/