Proposal Explorer

Add claims to see which ones the mechanisms address.

Operational use

• Operational use Sampled inference recomputation for checking untrusted workers' activations against the declared model, prompt and precision

Built for an adversarial prover

Sampled inference recomputation

No new hardware needed

Safeguard attestation and Sampled inference recomputation

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.

• Safeguard attestation: Analysis
• Sampled inference recomputation: Analysis

Open significant failures

4 failures in 2 mechanisms

• Components outside the attested boundary in Safeguard attestation

  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. ^{1 2}

  Known failure · Theoretical argument · Significant · Open. On the record

• Memory-bus interposition extracts attestation keys and forges attestations in Safeguard attestation

  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. ^{1 4 7 8 9 10 11 12 13}

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

  Known failure · Demonstrated attack · Significant · Open · Inherited finding. On the record · Related finding in TEE remote attestation for AI workloads

  Related mechanism Proposed Hardware-enabled guarantees (flexHEG) and guarantee processors: A tamper-protected enclosure around the chip is the proposed answer when the party that holds the hardware may attack it physically. A pointer, not evidence that this failure is mitigated. Add

• Tolerance for numerical noise leaves a covert channel in Sampled inference recomputation

  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. ^{14 22 25}

  Known failure · Demonstrated attack · Significant · Open. On the record

  Related mechanism Operational use Deterministic and bit-exact inference: Bit-exact inference would remove the numerical tolerance if exact replay can be deployed with the required weights and configuration. A pointer, not evidence that this failure is mitigated. Add

• Some inference optimizations are not covered in Sampled inference recomputation

  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. ^{15 16}

  Known failure · Theoretical argument · Significant · Open. On the record

Scope limitations

• Attestation shows a safeguard ran, not that it is effective in Safeguard attestation

  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. ¹

  Scope limitation · Theoretical argument. On the record

• Selective attestation leaves traffic uncovered in Safeguard attestation

  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. ^{1 4 9}

  Scope limitation · Theoretical argument. On the record

  Related mechanism Research demo On-chip telemetry from timing, memory and performance counters: On-chip counters are a proposed route to evidence about everything a chip runs, including a second virtual machine that skips the safeguard. A pointer, not evidence that this failure is mitigated. Add

• Measurements may omit behaviour-relevant configuration or runtime changes in Safeguard attestation

  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. ⁹

  Scope limitation · Theoretical argument. On the record

• Only recorded traffic is checked in Sampled inference recomputation

  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. ^{14 26}

  Scope limitation · Theoretical argument. On the record

  Related mechanism Proposed Network taps and certifiers: 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. A pointer, not evidence that this failure is mitigated. Add

Open minor failures

1 mechanism with minor failures

• Mixed hardware widens the honest baseline in Sampled inference recomputation

  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. ¹⁵

  Known failure · Open question · Minor · Open. On the record

Mechanisms on the map that are not in the proposal. Pointers, not recommendations. ?

• Operational use Deterministic and bit-exact inference Add
  • Bears on the open significant failure “Tolerance for numerical noise leaves a covert channel” in Sampled inference recomputation. Bit-exact inference would remove the numerical tolerance if exact replay can be deployed with the required weights and configuration.
• Proposed Hardware-enabled guarantees (flexHEG) and guarantee processors Add
  • 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.
• Operational use Hardware-attested weight binding Add
  • Safeguard attestation waits on it. Safeguard evidence must be bound to the model actually served, which depends on model-identity attestation.
• Operational use TEE remote attestation for AI workloads Add
  • 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.
• Proposed Network taps and certifiers Add
  • Sampled inference recomputation waits on it. In tap-based retrofit designs, recording all inference traffic needs network taps and recomputation servers that can ingest it, in the worst case one recomputation-server network interface per inference front-end interface.

Missing prerequisites

• Operational use TEE remote attestation for AI workloads needed by Safeguard attestation Add
• Operational use Hardware-attested weight binding needed by Safeguard attestation Add

Blockers

10 blockers recorded

• 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}
  • 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 27}
  • 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 evidence must be bound to the model actually served, which depends on model-identity attestation. Evidence binding. Waits on Hardware-attested weight binding ^{27 28}
  • 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}
• Sampled inference recomputation
  • In tap-based retrofit designs, recording all inference traffic needs network taps and recomputation servers that can ingest it, in the worst case one recomputation-server network interface per inference front-end interface. Coverage & hidden compute. Waits on Network taps and certifiers ^{21 29}
  • In retrofit designs, the recomputation server must sit inside the prover's data centre, possibly under the prover's physical control, and still be protected from a compromised provider, which Amodo rates 'not on track'. Hardware trust ^{14 21}
  • No independent red-team of a recomputation consistency check has been published (the one independent attack study targets the weight-exfiltration bound), and Amodo rates recomputation red-teaming 'not started'. Adversarial validation ^{21 22}
  • Tolerance-based checks need calibration on trusted hardware and exact knowledge of the provider's sampling procedure, and in one prototype a sampling-implementation mismatch produced large spurious differences. Performance & compatibility ^{15 20}
  • The verifier needs the model weights, so checking a closed-weights model requires a trusted, confidential recomputation environment, which the retrofit designs place inside the prover's facility. Privacy & leakage ^{13 15 26}

From the family or selected implementation's record. ?

Model weights

Depends on the design for Safeguard attestation and Sampled inference recomputation.

Inputs and outputs

Depends on the design for Safeguard attestation and Sampled inference recomputation.

Training data

Not involved: Safeguard attestation and Sampled inference recomputation.

Safeguard attestation

None on the map

Sampled inference recomputation

• Proposed AI 2040 inference-only verification stack Proposed architecture, AI Futures Project
• Research demo DiFR (Divergence From Reference) Research prototype
• Proposed Low-trust AI compute verification system overview Proposed architecture, Machine Intelligence Research Institute
• Proposed SASH confidential network logger Research prototype, Singapore AI Safety Hub (SASH)
• Operational use TOPLOC Open-source project, Prime Intellect

1. Proof-of-Guardrail in AI Agents and What (Not) to Trust from It, X. Jin et al. (2026). Original
2. Verifiable-ClawGuard: proof-of-guardrail reference code, SaharaLabsAI (2026). Original
3. Safety Without Compromising on Privacy, D. McCann-Sayles et al. (2026). Original
4. PAL*M: Property Attestation for Large Generative Models, P. Chantasantitam et al. (2026). Original
5. Enabling Verifiably-Scoped Monitoring through Large Language Models and Trusted Compute, B. Penchas et al. (2026). Original
6. Auditor-in-a-Box: Tools for Third-Party Auditing, R. Rinberg & B. Penchas (2026). Original
7. TEE.fail: Breaking Trusted Execution Environments via DDR5 Memory Bus Interposition, J. Chuang et al. (2026). Original
8. DDRop: Active Memory Interposer Attacks on Confidential VMs by Dropping DDR5 Writes, J. De Meulemeester et al. (2026). Original
9. On TEEs for Privacy-Preserving Monitoring in AI Governance, Gloria Z (2026). Original
10. Battering RAM: Low-Cost Interposer Attacks on Confidential Computing via Dynamic Memory Aliasing, J. De Meulemeester et al. (2026). Original
11. RMPocalypse: How a Catch-22 Breaks AMD SEV-SNP, B. Schlüter & S. Shinde (2025). Original
12. SEV-SNP RMP Initialization Vulnerability (AMD-SB-3020), AMD (2025). Original
13. A System Overview for Near-Term, Low-Trust AI Compute Verification, N. Cankaya (2026). Original
14. Verifying LLM Inference to Detect Model Weight Exfiltration, R. Rinberg et al. (2025). Original
15. DiFR: Inference Verification Despite Nondeterminism, A. Karvonen et al. (2025). Original
16. TOPLOC: A Locality Sensitive Hashing Scheme for Trustless Verifiable Inference, J. M. Ong et al. (2025). Original
17. PrimeIntellect-ai/toploc (GitHub repository), Prime Intellect (2025). Original
18. INTELLECT-2: A Reasoning Model Trained Through Globally Decentralized Reinforcement Learning, Prime Intellect Team et al. (2025). Original
19. adamkarvonen/difr (GitHub repository), A. Karvonen (2025). Original
20. Scaling Recomputation Inference Verification, Amodo Design (2026). Original
21. AI 2040 Plan A — Verification SITREP, Amodo Design (2026). Original
22. Adversarial Entropy Inflation Against Gumbel-Based Inference Verification, N. Kezins (2026). Original
23. SYNTHETIC-2 Release: Four Million Collaboratively Generated Reasoning Traces, Prime Intellect (2025). Original
24. An Inference Verification Prototype — Stage 1, Amodo Design (2026). Original
25. Bit-Exact AI Inference Verification Without Performance Tradeoffs, N. Cankaya (2026). Original
26. Example Schemes for Verifying High-Stakes AI Agreements, Amodo Design (2026). Original
27. Attestable Audits: Verifiable AI Safety Benchmarks Using Trusted Execution Environments, C. Schnabl et al. (2025). Original
28. How Tinfoil Proves Exactly What Model Is Running, Tinfoil Team (2026). Original
29. Verification Plan, R. Dean (2026). Original