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
• Operational use Deterministic and bit-exact inference for reproducing open-model inference from receipts in Gensyn's information-market service

Built for an adversarial prover

Sampled inference recomputation and Deterministic and bit-exact inference

No new hardware needed

Sampled inference recomputation, Deterministic and bit-exact inference and Chip registries and manufacturing records

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.

• Sampled inference recomputation: Analysis
• Deterministic and bit-exact inference: Analysis
• Chip registries and manufacturing records: Analysis

Open significant failures

3 failures in 2 mechanisms

• 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. ^{1 9 12}

  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. In the proposal.

• 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. ^{2 3}

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

• Documents and serial numbers can be forged in Chip registries and manufacturing records

  Avellar and Grunewald note that export documents can be forged, that companies can hide information behind obscure corporate structures, and that it may be possible to forge serial numbers on chips and racks. They recommend cryptographic attestation of a powered-on chip as an extra check. ²⁵

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

Scope limitations

• 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. ^{1 13}

  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

• Some kernels remain genuinely nondeterministic in Deterministic and bit-exact inference

  The bit-exact work separates kernels that are deterministic but not batch-invariant from truly nondeterministic ones that use atomic functions. Some integer de-quantization kernels use atomic additions and remain nondeterministic, so exact replay needs backends that avoid them. ¹²

  Scope limitation · Open question. On the record

• Cross-hardware replay relies on reverse-engineered, closed behaviour in Deterministic and bit-exact inference

  Emulating one GPU's rounding on another requires reverse-engineering tensor-core arithmetic and modelling proprietary kernel choices. Hawkeye covers a subset of NVIDIA architectures and states that attention and other higher-level operations need further reverse engineering. For the bit-exact emulator, a proprietary Hopper kernel family is an open edge case. ^{12 16}

  Scope limitation · Open question. On the record

• Records cover only chips that were recorded in Chip registries and manufacturing records

  A registry or commitment accounts only for chips entered into it. Cankaya asks how a verifier would know it had found all chips, or how much "dark compute" remains, and notes that a fraudulent original record would mean unregistered chips had been made in advance. Halstead and Larsen propose reconstructing earlier production by auditing upstream suppliers. ^{26 28}

  Scope limitation · Theoretical argument. On the record

  Related mechanism Proposed Remote detection of data centres: Looks for large data centres that were never declared, which a registry cannot show. A pointer, not evidence that this failure is mitigated. Add

• Insiders could alter records before they are fixed in Chip registries and manufacturing records

  Cankaya argues that insiders who can photograph process secrets could also tamper with production records. A commitment makes changes after publication detectable, but it cannot show that the records were accurate when committed. ²⁶

  Scope limitation · Theoretical argument. On the record

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. ²

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

Not yet demonstrated

Proposed Chip registries and manufacturing records

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

• 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.

Blockers

12 blockers recorded

• 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 ^{8 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 ^{1 8}
  • 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 ^{8 9}
  • 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 ^{2 7}
  • 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 ^{2 13 14}
• Deterministic and bit-exact inference
  • Batch-invariant kernels cost throughput: in Thinking Machines' Qwen3-8B test, an improved deterministic build took 42 s against 26 s for vLLM's default, and SGLang reports an average 34.35% slowdown on its FlashInfer and FlashAttention 3 backends. Performance & compatibility ^{15 17}
  • Coverage is incomplete: the bit-exact emulator targets dense blocks on NVIDIA GPUs and excludes mixture-of-experts inference and training, and vLLM's batch-invariant mode is in beta, with open work on AMD hardware and speculative decoding. Performance & compatibility ^{12 18 30}
  • Amodo's status page for the AI 2040 verification plan rates a reproducible inference stack for that plan as 'not started'. Performance & compatibility ⁸
  • Exact replay requires the prover to disclose weights, software versions, parallelism and batch sizes to whoever recomputes. Privacy & leakage ^{12 14}
• Chip registries and manufacturing records
  • No implementation of an AI chip registry has been publicly reported, and covering re-exports would need cooperation from re-exporters and foreign governments that may not be feasible everywhere. Access & governance ^{24 25}
  • Linking records to physical chips needs hard-to-spoof unique IDs and inspections. Hardware trust ^{24 25 26}
  • Chips produced before a registry starts must be reconstructed from supplier records. Coverage & hidden compute ^{26 28}

From the family or selected implementation's record. ?

Model weights

Depends on the design for Sampled inference recomputation and Deterministic and bit-exact inference.

Not involved: Chip registries and manufacturing records.

Inputs and outputs

Depends on the design for Sampled inference recomputation and Deterministic and bit-exact inference.

Not involved: Chip registries and manufacturing records.

Training data

Not involved: Sampled inference recomputation, Deterministic and bit-exact inference and Chip registries and manufacturing records.

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

Deterministic and bit-exact inference

• Research demo Batch-invariant inference kernels (Thinking Machines) Open-source project, Thinking Machines Lab
• Operational use Verde and RepOps (Gensyn) Product, Gensyn
• Proposed Low-trust AI compute verification system overview Proposed architecture, Machine Intelligence Research Institute

Chip registries and manufacturing records

None on the map

1. Verifying LLM Inference to Detect Model Weight Exfiltration, R. Rinberg et al. (2025). Original
2. DiFR: Inference Verification Despite Nondeterminism, A. Karvonen et al. (2025). Original
3. TOPLOC: A Locality Sensitive Hashing Scheme for Trustless Verifiable Inference, J. M. Ong et al. (2025). Original
4. PrimeIntellect-ai/toploc (GitHub repository), Prime Intellect (2025). Original
5. INTELLECT-2: A Reasoning Model Trained Through Globally Decentralized Reinforcement Learning, Prime Intellect Team et al. (2025). Original
6. adamkarvonen/difr (GitHub repository), A. Karvonen (2025). Original
7. Scaling Recomputation Inference Verification, Amodo Design (2026). Original
8. AI 2040 Plan A — Verification SITREP, Amodo Design (2026). Original
9. Adversarial Entropy Inflation Against Gumbel-Based Inference Verification, N. Kezins (2026). Original
10. SYNTHETIC-2 Release: Four Million Collaboratively Generated Reasoning Traces, Prime Intellect (2025). Original
11. An Inference Verification Prototype — Stage 1, Amodo Design (2026). Original
12. Bit-Exact AI Inference Verification Without Performance Tradeoffs, N. Cankaya (2026). Original
13. Example Schemes for Verifying High-Stakes AI Agreements, Amodo Design (2026). Original
14. A System Overview for Near-Term, Low-Trust AI Compute Verification, N. Cankaya (2026). Original
15. Defeating Nondeterminism in LLM Inference, H. He & Thinking Machines Lab (2025). Original
16. Hawkeye: Reproducing GPU-Level Non-Determinism, E. Badash et al. (2026). Original
17. Towards Deterministic Inference in SGLang and Reproducible RL Training, The SGLang Team (2025). Original
18. Batch Invariance (vLLM documentation), vLLM project (2026). Original
19. gensyn-ai/ree: Gensyn Reproducible Execution Environment (GitHub repository), Gensyn (2026). Original
20. EigenCloud Brings Verifiable AI to Mass Market with EigenAI and EigenCompute Launches, EigenCloud (2025). Original
21. Building Delphi: Pricing, Settlement, and Agentic Trading, D. Jedamski (2026). Original
22. Reproducible Execution Environment (REE) (Gensyn documentation), Gensyn (2026). Original
23. What is Delphi? (Delphi documentation), Gensyn (2026). Original
24. Verifying International Agreements on AI: Six Layers of Verification for Rules on Large-Scale AI Development and Deployment, M. Baker et al. (2025). Original
25. Near-Term Verification Methods for AI Chip Exports, B. Avellar & E. Grunewald (2026). Original
26. TSMC most definitely has a golden record of all AI chips it made, N. Cankaya (2025). Original
27. Hardware-Level Governance of AI Compute: A Feasibility Taxonomy for Regulatory Compliance and Treaty Verification, S. Ansari (2026). Original
28. Covert AI Projects, B. Halstead & T. Larsen (2026). Original
29. Verification Plan, R. Dean (2026). Original
30. [Feature]: Batch Invariant Feature and Performance Optimization (vLLM issue #27433), vLLM project contributors (2025). Original