# 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-0023,M-0016&implementations=M-0016:I-0021

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 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Safeguard attestation | Research demonstration | Published security analysis | Semi-trusted | Analysis | Existing features | 0 / 2 / 0 | depends | depends | not involved |
| Timed challenge-response and memory-occupation challenges / SAGE | Research demonstration | Published security analysis | Adversarial | Analysis | Existing features | 0 / 0 / 0 | unspecified | unspecified | unspecified |

## Claims

No claims chosen.

## Mechanisms

### Safeguard attestation

Hardware-signed evidence that an AI service sent a given response through its declared safeguard path, such as a wrapper that calls a guardrail classifier. ([Safeguard attestation](https://trustbutveri.fyi/mechanisms/safeguard-attestation/))

- Assessment: mechanism family.
- Development: Research demonstration (legacy code R2), assessed for attesting that a declared safeguard mediated a service's responses.
- Security evidence: Published security analysis. Independent evaluation: unassessed. Formal proof: unassessed. Deployment assurance: unassessed.
- Claims in this proposal: none of them.
- Threat model: semi-trusted prover. Hardware: existing features. Prover cooperation: required. Attack testing: analysis. Category: Cryptographic & computational.
- What the verifier sees: model weights depends; inputs and outputs depends; training data not involved. The enclave route signs hashes of the safeguard, request and response; a low-trust design has the verifier re-run and screen sampled requests itself.

### Timed challenge-response and memory-occupation challenges

A software-only attestation scheme in which a GPU must compute a checksum over its own code within a time limit, without trusted GPU hardware. ([Timed challenge-response and memory-occupation challenges](https://trustbutveri.fyi/mechanisms/timed-challenge-response/))

- Assessment: selected implementation [SAGE](https://trustbutveri.fyi/implementations/sage-gpu-attestation/).
- Development: Research demonstration (legacy code R2), assessed for attesting code execution on a GPU that lacks hardware trusted-execution support.
- 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: existing features. 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.


## Properties

**Built for an adversarial prover**

- Timed challenge-response and memory-occupation challenges

**No new hardware needed**

- Safeguard attestation
- Timed challenge-response and memory-occupation challenges


## 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**

- Safeguard attestation: Analysis
- Timed challenge-response and memory-occupation challenges / SAGE: Analysis


## Limits

**Open significant failures**

- Components outside the attested boundary (known failure, theoretical argument, in Safeguard attestation; https://trustbutveri.fyi/mechanisms/safeguard-attestation/evidence/flaws/4/) [1][2]. In the proof-of-guardrail experiments, the guardrail model and the agent's backend model were both reached through external APIs, and the authors leave the decision to trust those APIs to the verifier. The measured wrapper must also have no vulnerability that lets the unmeasured agent bypass the guardrail, for example by executing arbitrary commands inside the enclave. The code's README states that the enclave does not currently restrict the agent's arbitrary command execution, which could be used to bypass guardrails.
- Memory-bus interposition extracts attestation keys and forges attestations (known failure, demonstrated attack, in Safeguard attestation; https://trustbutveri.fyi/mechanisms/safeguard-attestation/evidence/flaws/5/) [1][4][7][8][9][10][11][12][13]. Inherited finding. Applies to variants using the affected Intel or AMD trust roots. PAL*M excludes physical attacks. A TDX-backed safeguard claim against a physical host attacker would be defeated, but these studies do not demonstrate a break of the AWS Nitro proof-of-guardrail prototype or of verifier-side recomputation. The TEE findings cover DDR5 attacks on Intel TDX, the H100 relay demonstration, DDR4 attacks on AMD SEV-SNP, and software-only SEV-SNP forgery before AMD's fixes. These are inherited hardware limits; a governance analysis explains why physical access matters in a treaty setting. Related finding: https://trustbutveri.fyi/mechanisms/tee-remote-attestation/evidence/flaws/1/.

  Response: Intel and AMD place the physical attack class outside their threat models, according to the researchers. AMD reports firmware fixes for RMPocalypse.

  Related mechanism: Hardware-enabled guarantees (flexHEG) and guarantee processors (R1, not in the proposal). A tamper-protected enclosure around the chip is the proposed answer when the party that holds the hardware may attack it physically.

**Family finding context**

- Context for SAGE. 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. Timing-based software attestation has been broken in practice (known failure, demonstrated attack, in Timed challenge-response and memory-occupation challenges; https://trustbutveri.fyi/mechanisms/timed-challenge-response/evidence/flaws/1/) [15][16]. Castelluccia et al. implemented two generic attacks, one based on a return-oriented rootkit and one on code compression, together with specific attacks on SWATT and ICE-based schemes, on commodity sensor nodes. They conclude that secure time-based attestation is "very difficult, if not impossible, to design correctly". The attacks target embedded schemes, not AI accelerators.

  Response: Perrig and van Doorn, two of the designers of SWATT and ICE, replied in August 2010. They argue that the rootkit attack defeats a naive implementation, not a property the schemes claim, and that the SWATT attack was run on a re-implementation on a chip with eight times the program memory, where SWATT's own chip is almost always full of code. They accept that the attack on ICE works.
- Context for SAGE. 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. Remote memory narrows the timing margin (known failure, theoretical argument, in Timed challenge-response and memory-occupation challenges; https://trustbutveri.fyi/mechanisms/timed-challenge-response/evidence/flaws/2/) [13]. Data-centre remote memory access returns in about 1–2 µs, against about 70–200 ns for local DRAM. The MIRI overview says verification of memory saturation depends on ruling out remote access by latency or physical disconnection. It names pre-staging data into local memory as the remaining evasion and proposes an unpredictable, capacity-filling challenge to close it.

  Related mechanism: Bandwidth limits and compartmentalization (R2, not in the proposal). Physical disconnection is proposed to exclude remote memory between the separated groups during a challenge. It depends on the isolation boundary being enforced.
- Context for SAGE. 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. Error rates not quantified (open question, open question, in Timed challenge-response and memory-occupation challenges; https://trustbutveri.fyi/mechanisms/timed-challenge-response/evidence/flaws/3/) [17]. Monfared et al. show separable timing distributions. Their acceptance rule passes a GPU when its mean time per round stays at or below a chosen maximum, and an appendix outlines statistical tests for the proof-of-work puzzle. They leave hardware-specific threshold values to future work and report no false-positive or false-negative rates.

**Scope limitations**

- Attestation shows a safeguard ran, not that it is effective (scope limitation, theoretical argument, in Safeguard attestation; https://trustbutveri.fyi/mechanisms/safeguard-attestation/evidence/flaws/1/) [1]. Proof of guardrail ensures that the guardrail executed, but the guardrail can still err or be jailbroken. Because the guardrail must be open source, a malicious developer can attack it with jailbreaks while still presenting a valid proof. In the authors' evaluation, Llama Guard 3 reached an F1 score of 0.56 on the unsafe class of the ToxicChat dataset. The authors state that proof of guardrail should not be interpreted or advertised as proof of safety.
- Selective attestation leaves traffic uncovered (scope limitation, theoretical argument, in Safeguard attestation; https://trustbutveri.fyi/mechanisms/safeguard-attestation/evidence/flaws/2/) [1][4][9]. Attestations are issued per response. In the prototype, the agent offers them when it receives high-stakes questions, so nothing shows that unattested traffic went through the same path. PAL*M's authors note that a prover could cherry-pick favourable executions, and suggest verifier-published nonces or requesting only session-level proofs. A governance analysis notes that auditors also need assurance that all activity is accounted for, since a host could start a second confidential virtual machine that bypasses monitoring.

  Related mechanism: On-chip telemetry from timing, memory and performance counters (R2, not in the proposal). On-chip counters are a proposed route to evidence about everything a chip runs, including a second virtual machine that skips the safeguard.
- Measurements may omit behaviour-relevant configuration or runtime changes (scope limitation, theoretical argument, in Safeguard attestation; https://trustbutveri.fyi/mechanisms/safeguard-attestation/evidence/flaws/3/) [9]. Every component that influences inference behaviour must be covered by the launch measurement, including feature flags, environment variables and invocation arguments. A launch measurement also does not show that a program keeps running as measured if the kernel is later compromised.
- Self-modifying code limits the timing margin (scope limitation, open question, in SAGE; https://trustbutveri.fyi/implementations/sage-gpu-attestation/evidence/flaws/1/) [14]. The authors report that their implementation reaches 75% of maximum GPU utilisation when the checksum uses self-modifying code, and that this limits the time difference caused by an adversary who inserts instructions into the checksum loop. They note that other cache-eviction strategies could raise utilisation.


## 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.

- **Hardware-enabled guarantees (flexHEG) and guarantee processors** (Proposed (legacy code R1), assessed for checking and enforcing training-compute limits on chips, against adversaries up to states)
  - Bears on the open significant failure "Memory-bus interposition extracts attestation keys and forges attestations" in Safeguard attestation. A tamper-protected enclosure around the chip is the proposed answer when the party that holds the hardware may attack it physically.
- **Hardware-attested weight binding** (Operational use (legacy code R3), assessed for hardware-attested weight binding showing users that a service runs its committed weights)
  - Safeguard attestation waits on it: Safeguard evidence must be bound to the model actually served, which depends on model-identity attestation.
- **TEE remote attestation for AI workloads** (Operational use (legacy code R3), assessed for showing which software ran to a party that distrusts the operator holding the hardware)
  - Safeguard attestation waits on it: Frontier model inference typically needs several GPUs, GPU confidential computing is less mature than CPU support, and CPU inference, which an enclave prototype had to use, ran about 100 times slower than GPU inference.
  - Safeguard attestation waits on it: Trust rests on a small number of hardware vendors, and a per-CPU Intel attestation key has been extracted by physical attack.


## Dependencies

**Missing prerequisites**

- TEE remote attestation for AI workloads (Operational use (legacy code R3), assessed for showing which software ran to a party that distrusts the operator holding the hardware), needed by Safeguard attestation
- Hardware-attested weight binding (Operational use (legacy code R3), assessed for hardware-attested weight binding showing users that a service runs its committed weights), needed by Safeguard attestation

**Blockers**

- Safeguard attestation: No published design shows that all of a provider's traffic passes through the attested safeguard path; current evidence covers individual attested responses. (coverage & hidden compute) [1][9]
- Safeguard attestation: Frontier model inference typically needs several GPUs, GPU confidential computing is less mature than CPU support, and CPU inference, which an enclave prototype had to use, ran about 100 times slower than GPU inference. (performance & compatibility; waits on TEE remote attestation for AI workloads) [9][18]
- Safeguard attestation: Trust rests on a small number of hardware vendors, and a per-CPU Intel attestation key has been extracted by physical attack. (hardware trust; waits on TEE remote attestation for AI workloads) [7][9]
- Safeguard attestation: Safeguard evidence must be bound to the model actually served, which depends on model-identity attestation. (evidence binding; waits on Hardware-attested weight binding) [18][19]
- Safeguard attestation: No independent red-team or audit of a safeguard-attestation system has been published, and the available prototypes are described by their authors as proofs of concept that have not been stress-tested by a counterparty. (adversarial validation) [2][6]
- Timed challenge-response and memory-occupation challenges: The verifier must know the exact hardware configuration of the GPU. (hardware trust) [14]
- Timed challenge-response and memory-occupation challenges: The verifier runs in an SGX enclave on the same host as the GPU, so the scheme inherits trust in that enclave. (hardware trust) [14]


## What the verifier sees

- Model weights: shown by none; depends on the design for Safeguard attestation; hidden by none; not involved in none; unspecified for Timed challenge-response and memory-occupation challenges.
- Inputs and outputs: shown by none; depends on the design for Safeguard attestation; hidden by none; not involved in none; unspecified for Timed challenge-response and memory-occupation challenges.
- Training data: shown by none; depends on the design for none; hidden by none; not involved in Safeguard attestation; unspecified for Timed challenge-response and memory-occupation challenges.

## Implementations

- Safeguard attestation: none on the map
- Timed challenge-response and memory-occupation challenges: [Data-centre memory challenging](https://trustbutveri.fyi/implementations/data-centre-memory-challenging/) (R1, proposed architecture); [GPU contention probes](https://trustbutveri.fyi/implementations/gpu-contention-probes/) (R2, research prototype); [Low-trust AI compute verification system overview](https://trustbutveri.fyi/implementations/low-trust-compute-verification-system-overview/) (R1, proposed architecture); [SAGE](https://trustbutveri.fyi/implementations/sage-gpu-attestation/) (R2, research prototype); [VRAM-residency challenge](https://trustbutveri.fyi/implementations/vram-residency-challenge/) (R2, research prototype)

## Sources

1. Proof-of-Guardrail in AI Agents and What (Not) to Trust from It, X. Jin et al. (2026). https://arxiv.org/abs/2603.05786
2. Verifiable-ClawGuard: proof-of-guardrail reference code, SaharaLabsAI (2026). https://github.com/SaharaLabsAI/Verifiable-ClawGuard
3. Safety Without Compromising on Privacy, D. McCann-Sayles et al. (2026). https://tinfoil.sh/blog/2026-09-14-safety-without-compromising-privacy
4. PAL*M: Property Attestation for Large Generative Models, P. Chantasantitam et al. (2026). https://arxiv.org/abs/2601.16199
5. Enabling Verifiably-Scoped Monitoring through Large Language Models and Trusted Compute, B. Penchas et al. (2026). https://icml.cc/virtual/2026/78630
6. Auditor-in-a-Box: Tools for Third-Party Auditing, R. Rinberg & B. Penchas (2026). https://www.lesswrong.com/posts/uWYk7MM9hAf9GEbGe/auditor-in-a-box-tools-for-third-party-auditing
7. TEE.fail: Breaking Trusted Execution Environments via DDR5 Memory Bus Interposition, J. Chuang et al. (2026). https://tee.fail/
8. DDRop: Active Memory Interposer Attacks on Confidential VMs by Dropping DDR5 Writes, J. De Meulemeester et al. (2026). https://ddropattack.eu/
9. On TEEs for Privacy-Preserving Monitoring in AI Governance, Gloria Z (2026). https://techgov.intelligence.org/blog/on-tees-for-privacy-preserving-monitoring-in-ai-governance
10. Battering RAM: Low-Cost Interposer Attacks on Confidential Computing via Dynamic Memory Aliasing, J. De Meulemeester et al. (2026). https://batteringram.eu/
11. RMPocalypse: How a Catch-22 Breaks AMD SEV-SNP, B. Schlüter & S. Shinde (2025). https://rmpocalypse.github.io/
12. SEV-SNP RMP Initialization Vulnerability (AMD-SB-3020), AMD (2025). https://www.amd.com/en/resources/product-security/bulletin/amd-sb-3020.html
13. 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
14. SAGE: Software-based Attestation for GPU Execution, A. Ivanov et al. (2023). https://www.usenix.org/conference/atc23/presentation/ivanov
15. On the Difficulty of Software-Based Attestation of Embedded Devices, C. Castelluccia et al. (2009). https://s3.eurecom.fr/docs/ccs09_Castelluccia.pdf
16. Refutation of "On the Difficulty of Software-Based Attestation of Embedded Devices", A. Perrig & L. van Doorn (2010). https://netsec.ethz.ch/publications/papers/perrig-ccs-refutation.pdf
17. Timing and Memory Telemetry on GPUs for AI Governance, S. K. Monfared et al. (2026). https://arxiv.org/abs/2602.09369
18. Attestable Audits: Verifiable AI Safety Benchmarks Using Trusted Execution Environments, C. Schnabl et al. (2025). https://arxiv.org/abs/2506.23706
19. How Tinfoil Proves Exactly What Model Is Running, Tinfoil Team (2026). https://tinfoil.sh/blog/2026-02-03-proving-model-identity