# 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-0014,M-0012&implementations=M-0012:I-0007

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 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Bandwidth limits and compartmentalization | Research demonstration | Published security analysis | Adversarial | Analysis | Retrofit device | 0 / 2 / 0 | not involved | not involved | not involved |
| Hardware-attested weight binding / Attestable Audits | Research demonstration | Published security analysis | Semi-trusted | Analysis | Existing features | 0 / 0 / 0 | unspecified | unspecified | unspecified |

## Claims

No claims chosen.

## Mechanisms

### Bandwidth limits and compartmentalization

Capping or removing network links between groups of AI chips, so each group can serve models but large training runs across groups become far slower. ([Bandwidth limits and compartmentalization](https://trustbutveri.fyi/mechanisms/bandwidth-limits-and-compartmentalization/))

- Assessment: mechanism family.
- Development: Research demonstration (legacy code R2), assessed for monitoring inter-node traffic with operator-run software on four GPUs.
- 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: Isolation & system architectures.
- What the verifier sees: model weights not involved; inputs and outputs not involved; training data not involved. Caps traffic between groups of chips; it does not read the traffic's content.

### Hardware-attested weight binding

A research prototype that runs AI safety benchmarks inside a trusted execution environment and publishes hardware-signed records linking the model, the audit and the results. ([Hardware-attested weight binding](https://trustbutveri.fyi/mechanisms/model-identity-attestation/))

- Assessment: selected implementation [Attestable Audits](https://trustbutveri.fyi/implementations/attestable-audits/).
- Development: Research demonstration (legacy code R2), assessed for showing users that the model answering them is the audited one.
- 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: On-chip & hardware-enabled.
- 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**

- Bandwidth limits and compartmentalization

**No new hardware needed**

- Hardware-attested weight binding


## 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**

- Bandwidth limits and compartmentalization: Analysis
- Hardware-attested weight binding / Attestable Audits: Analysis


## Limits

**Open significant failures**

- Operator control of pod routing collapses the bound (known failure, theoretical argument, in Bandwidth limits and compartmentalization; https://trustbutveri.fyi/mechanisms/bandwidth-limits-and-compartmentalization/evidence/flaws/2/) [2]. Lucid's analysis finds that if the operator can freely assign pods to routers, it could dedicate a whole cell of 100 or more pods to one pipeline stage. The bound then falls to about 90–220x uncompressed and as low as about 25x with compression. The proposed mitigation, auditor-controlled random assignment that is periodically re-randomized, has not been implemented.
- Parallel scale-up switches are hard enforcement points (known failure, theoretical argument, in Bandwidth limits and compartmentalization; https://trustbutveri.fyi/mechanisms/bandwidth-limits-and-compartmentalization/evidence/flaws/5/) [5]. In GB200 topologies, GPUs reach GPUs in other nodes through NVSwitches without a NIC on the path. Amodo notes that limits are hard to enforce there because many switches work in parallel, so compromising one or two would bypass the limit.

**Family finding context**

- Context for Attestable Audits. 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. Underlying attestation can be forged or relayed (known failure, demonstrated attack, in Hardware-attested weight binding; https://trustbutveri.fyi/mechanisms/model-identity-attestation/evidence/flaws/1/) [10][11][12][13][14][15]. Inherited finding. Critical for weight binding against an operator with physical access to affected hardware, or with control of the hypervisor on an AMD SEV-SNP platform without AMD's fixes. PAL*M excludes physical attacks, and Tinfoil acknowledges this boundary. The enclave route inherits the platform-specific TEE attestation failures. Intel TDX forgery and H100 relay were demonstrated with physical access and host control. Battering RAM defeated AMD SEV-SNP attestation on DDR4 servers; RMPocalypse did so from malicious host software on platforms without AMD's fixes. These demonstrate failures of the trust roots, not of each model-commitment protocol. Related finding: https://trustbutveri.fyi/mechanisms/tee-remote-attestation/evidence/flaws/1/.

  Response: The TEE.fail authors report that physical interposer attacks are outside Intel's and AMD's threat models. AMD reports 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.
- Context for Attestable Audits. 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. Launch-state attestation does not by itself cover weights loaded later (known failure, theoretical argument, in Hardware-attested weight binding; https://trustbutveri.fyi/mechanisms/model-identity-attestation/evidence/flaws/2/) [16][17]. Attestation measures launch state, and weights are read from disk after boot. A signature checked at load time does not stop a malicious hypervisor from altering the disk afterwards. Tinfoil reports mitigating this with dm-verity checks on every read. Unmeasured runtime configuration remains a general risk.
- Context for Attestable Audits. 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. For private models, a user can confirm consistency but not content (scope limitation, open question, in Hardware-attested weight binding; https://trustbutveri.fyi/mechanisms/model-identity-attestation/evidence/flaws/3/) [8][16]. When weights are not published, users can check that the same root hash is served each time, but not what the model is. Pairing the hash with an attested evaluation, as in Attestable Audits, is one proposed remedy.

**Scope limitations**

- Undeclared local storage raises per-pod capacity (scope limitation, theoretical argument, in Bandwidth limits and compartmentalization; https://trustbutveri.fyi/mechanisms/bandwidth-limits-and-compartmentalization/evidence/flaws/3/) [2]. More memory or storage per pod helps an adversary. Lucid requires per-pod storage to be declared, capped and physically inspected.
- Training within one pod is not covered (scope limitation, open question, in Bandwidth limits and compartmentalization; https://trustbutveri.fyi/mechanisms/bandwidth-limits-and-compartmentalization/evidence/flaws/4/) [2]. Lucid's bounds concern pre-training models larger than the pods are sized for. Training models that fit in one pod, fine-tuning and reinforcement-learning post-training within one pod are outside the modelled threat.
- Relies on the TEE vendor and inherits TEE attacks (scope limitation, theoretical argument, in Attestable Audits; https://trustbutveri.fyi/implementations/attestable-audits/evidence/flaws/1/) [8]. Inherited finding. The prototype trusts AWS Nitro, not the Intel TDX or AMD SEV-SNP attestation roots targeted by the cited confidential-VM studies. Those studies are class context, not a demonstrated attack on this Nitro prototype. The design depends on trusting the TEE vendor, AWS in the prototype. The authors cite memory-aliasing, ciphertext side-channel and malicious-interrupt attacks on confidential VMs (BadRAM, CIPHERLEAKS, Heckler). Their answer is to revoke vulnerable base images once such attacks are discovered. Related finding: https://trustbutveri.fyi/mechanisms/tee-remote-attestation/evidence/flaws/5/.

  Response: The authors propose revoking vulnerable base images; they do not report a red-team evaluation of the prototype.
- CPU-only enclaves force small, quantized models and high cost (scope limitation, open question, in Attestable Audits; https://trustbutveri.fyi/implementations/attestable-audits/evidence/flaws/3/) [8][9]. Memory limits required 4-bit quantization, and the quantized model scored 51.4% on zero-shot MMLU. CPU inference cost 21.7 times as much per token as GPU inference, and the enclave roughly doubled the CPU cost. The authors wrote that H100 confidential computing had no multi-GPU support. NVIDIA's white paper of August 2025 describes a protected-PCIe mode that passes all eight GPUs of a Hopper HGX node to one confidential VM, with NVLink traffic unencrypted.

**Open questions**

- Low-communication training reduces the bandwidth training needs (open question, theoretical argument, in Bandwidth limits and compartmentalization; https://trustbutveri.fyi/mechanisms/bandwidth-limits-and-compartmentalization/evidence/flaws/1/) [2][6][7]. DiLoCo matched fully synchronous training on 8 workers while communicating 500 times less. Rahman writes that this family of methods theoretically allows large-scale training with less than 100 Mbps. Lucid includes these methods in its bounds, but notes that extreme activation compression, architectures with unusually small inter-layer widths, or modular paradigms could erode the margin.
- Prompt-based model exfiltration is a residual gap (open question, open question, in Attestable Audits; https://trustbutveri.fyi/implementations/attestable-audits/evidence/flaws/2/) [8]. The authors state that "prompt-based model exfiltration during the user interaction step remains a residual gap".


## 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.

- **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)
  - Hardware-attested weight binding waits on it: The prototype needs porting to GPU confidential computing to handle larger models; the authors expect an overhead as small as 5 times there.
- **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)
  - Bandwidth limits and compartmentalization waits on it: Shaping devices and routing assignments must be trusted by both parties; Amodo has not yet fully analysed resilience to a compromised DPU.
- **Network taps and certifiers** (Proposed (legacy code R1), assessed for committing a complete record of cluster traffic, so declared inference can be checked)
  - Bandwidth limits and compartmentalization waits on it: The verifier must know that all traffic leaving a pod crosses the capped, monitored links.


## Dependencies

**Missing prerequisites**

- 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 Bandwidth limits and compartmentalization
- 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 Hardware-attested weight binding

**Blockers**

- Bandwidth limits and compartmentalization: No cap that a verifier can check has been implemented or red-teamed. (adversarial validation) [2]
- Bandwidth limits and compartmentalization: The verifier must know that all traffic leaving a pod crosses the capped, monitored links. (coverage & hidden compute; waits on Network taps and certifiers) [3]
- Bandwidth limits and compartmentalization: Shaping devices and routing assignments must be trusted by both parties; Amodo has not yet fully analysed resilience to a compromised DPU. (hardware trust; waits on Tamper evidence for verifier devices) [2][5]
- Bandwidth limits and compartmentalization: Advances in low-communication training could shrink the margin that the cap enforces. (capacity bounds) [2][6][7]
- Hardware-attested weight binding: The prototype needs porting to GPU confidential computing to handle larger models; the authors expect an overhead as small as 5 times there. (performance & compatibility; waits on TEE remote attestation for AI workloads) [8]
- Hardware-attested weight binding: As of September 2026 no code has been released for the prototype. (adversarial validation) [8]


## What the verifier sees

- Model weights: shown by none; depends on the design for none; hidden by none; not involved in Bandwidth limits and compartmentalization; unspecified for Hardware-attested weight binding.
- Inputs and outputs: shown by none; depends on the design for none; hidden by none; not involved in Bandwidth limits and compartmentalization; unspecified for Hardware-attested weight binding.
- Training data: shown by none; depends on the design for none; hidden by none; not involved in Bandwidth limits and compartmentalization; unspecified for Hardware-attested weight binding.

## Implementations

- Bandwidth limits and compartmentalization: [AI 2040 inference-only verification stack](https://trustbutveri.fyi/implementations/ai-2040-inference-only-verification-plan/) (R1, proposed architecture); [RAND secure inference data center (SIDC) design](https://trustbutveri.fyi/implementations/rand-secure-inference-data-centers/) (R1, proposed architecture)
- Hardware-attested weight binding: [Attestable Audits](https://trustbutveri.fyi/implementations/attestable-audits/) (R2, research prototype); [PAL*M](https://trustbutveri.fyi/implementations/palm/) (R2, research prototype); [Tinfoil model identity (Modelwrap)](https://trustbutveri.fyi/implementations/tinfoil-model-identity/) (R3, product)

## Sources

1. Verification Plan, R. Dean (2026). https://ai-2040.com/supplements/verification-plan
2. Traffic Shaping for Workload Classification, Lucid Computing (2026). https://lucidcomputing.substack.com/p/traffic-shaping-for-workload-classification
3. 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
4. De-risking Interconnect Limits for AI Verification, A. Scher et al. (2026). https://techgov.intelligence.org/blog/de-risking-interconnect-limits-for-ai-verification
5. The Tray as a Bandwidth Boundary, Amodo Design (2026). https://amododesign.com/notes/2026-03-16-dpu-bandwidth-limiter/
6. DiLoCo: Distributed Low-Communication Training of Language Models, A. Douillard et al. (2024). https://arxiv.org/abs/2311.08105
7. Does Distributed Training Undermine Compute Governance?, R. Rahman (2026). https://arxiv.org/abs/2605.29359
8. Attestable Audits: Verifiable AI Safety Benchmarks Using Trusted Execution Environments, C. Schnabl et al. (2025). https://arxiv.org/abs/2506.23706
9. NVIDIA Secure AI with Blackwell and Hopper GPUs (White Paper), NVIDIA (2025). https://docs.nvidia.com/nvidia-secure-ai-with-blackwell-and-hopper-gpus-whitepaper.pdf
10. TEE.fail: Breaking Trusted Execution Environments via DDR5 Memory Bus Interposition, J. Chuang et al. (2026). https://tee.fail/
11. Battering RAM: Low-Cost Interposer Attacks on Confidential Computing via Dynamic Memory Aliasing, J. De Meulemeester et al. (2026). https://batteringram.eu/
12. RMPocalypse: How a Catch-22 Breaks AMD SEV-SNP, B. Schlüter & S. Shinde (2025). https://rmpocalypse.github.io/
13. SEV-SNP RMP Initialization Vulnerability (AMD-SB-3020), AMD (2025). https://www.amd.com/en/resources/product-security/bulletin/amd-sb-3020.html
14. A primer on secure enclaves, Tinfoil (2026). https://docs.tinfoil.sh/verification/secure-enclave-primer
15. PAL*M: Property Attestation for Large Generative Models, P. Chantasantitam et al. (2026). https://arxiv.org/abs/2601.16199
16. How Tinfoil Proves Exactly What Model Is Running, Tinfoil Team (2026). https://tinfoil.sh/blog/2026-02-03-proving-model-identity
17. 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