{ "schema_version": "1.3", "url": "https://trustbutveri.fyi/explorer/?mechanisms=M-0023,M-0024,M-0010,M-0001", "data_generated": "2026-10-09", "definitions": { "methodology": "https://trustbutveri.fyi/about/methodology/", "readiness": "https://trustbutveri.fyi/about/readiness/", "filters": [ { "id": "prover", "label": "Prover", "question": "How far can the party being checked be trusted?", "options": [ { "value": "cooperative", "label": "Cooperative" }, { "value": "semi-trusted", "label": "Semi-trusted" }, { "value": "adversarial", "label": "Adversarial" } ], "rule": "Keeps mechanisms whose threat model holds against at least this prover. Adversarial is the strongest assumption.", "about": "The prover is the party being checked. Semi-trusted designs rely on part of its stack: usually the chip vendor's hardware root of trust, its firmware or counters, or its supply-chain records. Adversarial designs aim to hold even if it cheats wherever the checks allow, within their stated assumptions." }, { "id": "onsite", "label": "Verifier devices on site", "question": "May the verifier install its own hardware at the prover's sites?", "options": [ { "value": "no", "label": "Not allowed" } ], "rule": "\"Not allowed\" removes mechanisms that need a retrofit device, such as a network tap or a sealed sensor.", "about": "Some mechanisms need a device the verifier owns or trusts at the prover's facility, such as a network tap, a bandwidth limiter or a sealed sensor. Choose Not allowed when the setting rules that out. Inspectors are not covered." }, { "id": "coop", "label": "Prover cooperation", "question": "How much must the prover take part?", "options": [ { "value": "partial", "label": "Partial at most" }, { "value": "none", "label": "Not required" } ], "rule": "\"Partial at most\" removes mechanisms that need the prover's active participation. \"Not required\" keeps only those that work without it.", "about": "Required: the prover takes part, for example by logging requests, producing proofs or opening records. Partial: some access, such as installing a device. Not required: works from outside, such as satellite imagery." }, { "id": "chips", "label": "Chips", "question": "May the proposal depend on new chip designs?", "options": [ { "value": "existing", "label": "Existing chips only" } ], "rule": "\"Existing chips only\" removes mechanisms that need changes to future chip designs.", "about": "New chip features take years to reach a deployed fleet and cover only chips made after they ship. Mechanisms that use shipping features, such as trusted execution environments or performance counters, stay." }, { "id": "ready", "label": "Minimum development status", "question": "Development status", "options": [ { "value": "R1", "label": "Proposed" }, { "value": "R2", "label": "Research demonstration" }, { "value": "R3", "label": "Operational use" }, { "value": "R4", "label": "Legacy independent-evaluation filter", "legacy": true } ], "rule": "Keeps mechanisms whose readiness level is at least this one.", "about": "A level describes the public evidence for a mechanism's stated use, not its cost or feasibility. R3 can still have open critical flaws." }, { "id": "tested", "label": "Attack testing", "question": "How hard has each mechanism been attacked in public?", "options": [ { "value": "analysis", "label": "Published analysis" }, { "value": "red-teamed", "label": "Red-teamed" }, { "value": "independent-red-team", "label": "Independent red-team" } ], "rule": "Keeps mechanisms whose strongest published attack testing is at least this.", "about": "The strongest published attempt to break the mechanism for its verification use: a security analysis, red-teaming by its developers or collaborators, or a red team independent of them." }, { "id": "hide", "label": "Keep hidden from the verifier", "question": "What must the verifier never see?", "options": [ { "value": "weights", "label": "Model weights" }, { "value": "io", "label": "Inputs and outputs" }, { "value": "training", "label": "Training data" } ], "rule": "Removes mechanisms that show the asset to the verifier. Conditional or unspecified exposure stays with a note and needs checking against the privacy requirement.", "about": "Model weights: the checked model's parameters. Inputs and outputs: the requests a deployed model serves and its responses. Training data: what a model was trained on. Each mechanism's exposure is the editors' reading of its record: shown, depends on the design (kept, with a note), hidden, not involved, or unspecified for a selected implementation. Code and configuration are not covered yet." } ], "exposure": "For model weights, inputs and outputs, and training data. This is the editors' reading of each mechanism's record (its threat model, how it works and its limitations), not a field of the record. Shown: the verifier sees it. Depends: on the design or variant, or the verifier sees only samples. Hidden: the verifier sees only commitments, hashes, proofs or results. Not involved: the record does not handle it. Unspecified: the selected implementation has no asset-specific assessment here.", "claim_status": { "addressed": "A mechanism in the proposal is aimed at this claim and is not excluded by the filters.", "partly-addressed": "Only supporting mechanisms, or mechanisms aimed at it that the filters exclude.", "unaddressed": "No mechanism in the proposal addresses this claim." }, "finding_classification": { "failure": { "label": "Known failures", "singular": "Known failure", "anchor": "known-flaws" }, "scope-limitation": { "label": "Scope limitations", "singular": "Scope limitation", "anchor": "scope-limitations" }, "open-question": { "label": "Open questions", "singular": "Open question", "anchor": "open-questions" } }, "finding_scope": "Evidence scope describes where a finding was demonstrated; it does not establish applicability to every implementation in the mechanism family.", "claim_finding_scope": "open_critical_findings names active failures on the assessed records; open_critical_context names conditional family failures whose implementation applicability is unassessed.", "legacy_status": "The status field retains covered/partial/none for compatibility. It names claim links, never successful verification. Use claim_status and status_label for presentation." }, "filters": { "prover": "", "onsite": "", "coop": "", "chips": "", "ready": "", "tested": "", "hide": [] }, "mechanisms_passing_filters": 25, "claims": [], "mechanisms": [ { "id": "M-0023", "title": "Safeguard attestation", "url": "https://trustbutveri.fyi/mechanisms/safeguard-attestation/", "assessment_record": { "id": "M-0023", "title": "Safeguard attestation", "url": "https://trustbutveri.fyi/mechanisms/safeguard-attestation/" }, "finding_counts": { "failure": 2, "scope_limitation": 3, "open_question": 0, "open_failures": { "critical": 0, "significant": 2, "minor": 0 } }, "selected_implementation": null, "readiness": { "level": "R2", "scope": "attesting that a declared safeguard mediated a service's responses", "confidence": "low", "evidence": [ "S-1500", "S-1501", "S-3362", "S-0012", "S-1503", "S-1504", "S-1202", "S-3126" ] }, "development_status": { "code": "R2", "label": "Research demonstration", "short": "Research demo", "rank": 2, "legacy_code": "R2" }, "security_evidence": { "attack_testing": { "status": "analysis", "label": "Published security analysis", "kind": "analysis" }, "independent_evaluation": { "status": "unassessed" }, "formal_proof": { "status": "unassessed" }, "deployment_assurance": { "status": "unassessed" }, "legacy_evaluation_code": null, "scoped_findings": [ { "n": 4, "severity": "significant", "status": "open", "evidence_scope": "unassessed", "sources": [ "S-1500", "S-1501" ] }, { "n": 5, "severity": "significant", "status": "open", "evidence_scope": "inherited", "sources": [ "S-1202", "S-3126", "S-1210", "S-1212", "S-1213", "S-0012", "S-0014", "S-1500", "S-0018" ], "related_finding": { "record": "M-0008", "flaw": 1 } } ], "open_failures": { "critical": 0, "significant": 2, "minor": 0 } }, "assessed_properties": { "threat_model": "semi-trusted", "hardware_requirement": "existing-features", "prover_cooperation": "required", "adversarial_evaluation": "analysis" }, "claims": [], "exposure": { "weights": "partial", "io": "partial", "training": "none", "note": "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." }, "family_finding_context": [], "filter_issues": [] }, { "id": "M-0024", "title": "Bounding unexplained information in outputs", "url": "https://trustbutveri.fyi/mechanisms/bounding-unexplained-information/", "assessment_record": { "id": "M-0024", "title": "Bounding unexplained information in outputs", "url": "https://trustbutveri.fyi/mechanisms/bounding-unexplained-information/" }, "finding_counts": { "failure": 1, "scope_limitation": 2, "open_question": 1, "open_failures": { "critical": 0, "significant": 1, "minor": 0 } }, "selected_implementation": null, "readiness": { "level": "R2", "scope": "bounding how much hidden information can leave in checked inference outputs", "confidence": "low", "evidence": [ "S-0019", "S-0015", "S-1507" ] }, "development_status": { "code": "R2", "label": "Research demonstration", "short": "Research demo", "rank": 2, "legacy_code": "R2" }, "security_evidence": { "attack_testing": { "status": "independent-red-team", "label": "Published attack testing", "kind": "practical", "attribution": "Independent team" }, "independent_evaluation": { "status": "unassessed" }, "formal_proof": { "status": "unassessed" }, "deployment_assurance": { "status": "unassessed" }, "legacy_evaluation_code": null, "scoped_findings": [ { "n": 1, "severity": "significant", "status": "open", "evidence_scope": "unassessed", "sources": [ "S-1507", "S-0015" ] } ], "open_failures": { "critical": 0, "significant": 1, "minor": 0 } }, "assessed_properties": { "threat_model": "adversarial", "hardware_requirement": "retrofit-device", "prover_cooperation": "required", "adversarial_evaluation": "independent-red-team" }, "claims": [], "exposure": { "weights": "partial", "io": "partial", "training": "none", "note": "Depends on where recomputation runs: in a sealed enclosure, or with zero-knowledge proofs, the verifier need not see the weights or the traffic." }, "family_finding_context": [], "filter_issues": [] }, { "id": "M-0010", "title": "On-chip telemetry from timing, memory and performance counters", "url": "https://trustbutveri.fyi/mechanisms/on-chip-telemetry/", "assessment_record": { "id": "M-0010", "title": "On-chip telemetry from timing, memory and performance counters", "url": "https://trustbutveri.fyi/mechanisms/on-chip-telemetry/" }, "finding_counts": { "failure": 2, "scope_limitation": 2, "open_question": 1, "open_failures": { "critical": 0, "significant": 2, "minor": 0 } }, "selected_implementation": null, "readiness": { "level": "R2", "scope": "workload evidence from GPU counters and timing, assuming authentic measurements", "confidence": "medium", "evidence": [ "S-0033", "S-0034", "S-0037" ] }, "development_status": { "code": "R2", "label": "Research demonstration", "short": "Research demo", "rank": 2, "legacy_code": "R2" }, "security_evidence": { "attack_testing": { "status": "red-teamed", "label": "Published attack testing", "kind": "practical", "attribution": "Developers or collaborators" }, "independent_evaluation": { "status": "unassessed" }, "formal_proof": { "status": "unassessed" }, "deployment_assurance": { "status": "unassessed" }, "legacy_evaluation_code": null, "scoped_findings": [ { "n": 2, "severity": "significant", "status": "open", "evidence_scope": "unassessed", "sources": [ "S-0037" ] }, { "n": 4, "severity": "significant", "status": "open", "evidence_scope": "unassessed", "sources": [ "S-0014", "S-1200" ] } ], "open_failures": { "critical": 0, "significant": 2, "minor": 0 } }, "assessed_properties": { "threat_model": "semi-trusted", "hardware_requirement": "existing-features", "prover_cooperation": "partial", "adversarial_evaluation": "red-teamed" }, "claims": [], "exposure": { "weights": "partial", "io": "partial", "training": "partial", "note": "Counters do not read weights or data, but richer counters can leak secrets through side channels." }, "family_finding_context": [], "filter_issues": [] }, { "id": "M-0001", "title": "Sampled inference recomputation", "url": "https://trustbutveri.fyi/mechanisms/sampled-inference-recomputation/", "assessment_record": { "id": "M-0001", "title": "Sampled inference recomputation", "url": "https://trustbutveri.fyi/mechanisms/sampled-inference-recomputation/" }, "finding_counts": { "failure": 3, "scope_limitation": 1, "open_question": 0, "open_failures": { "critical": 0, "significant": 2, "minor": 1 } }, "selected_implementation": null, "readiness": { "level": "R3", "scope": "checking untrusted workers' activations against the declared model, prompt and precision", "confidence": "low", "evidence": [ "S-0015", "S-0016", "S-1000", "S-1001", "S-1003", "S-1005", "S-1006", "S-1008", "S-1507", "S-3000", "S-3001" ] }, "development_status": { "code": "R3", "label": "Operational use", "short": "Operational use", "rank": 3, "legacy_code": "R3" }, "security_evidence": { "attack_testing": { "status": "analysis", "label": "Published security analysis", "kind": "analysis" }, "independent_evaluation": { "status": "unassessed" }, "formal_proof": { "status": "unassessed" }, "deployment_assurance": { "status": "unassessed" }, "legacy_evaluation_code": null, "scoped_findings": [ { "n": 1, "severity": "significant", "status": "open", "evidence_scope": "unassessed", "sources": [ "S-0020", "S-0015", "S-1507" ] }, { "n": 3, "severity": "significant", "status": "open", "evidence_scope": "unassessed", "sources": [ "S-1000", "S-0016" ] }, { "n": 4, "severity": "minor", "status": "open", "evidence_scope": "unassessed", "sources": [ "S-0016" ] } ], "open_failures": { "critical": 0, "significant": 2, "minor": 1 } }, "assessed_properties": { "threat_model": "adversarial", "hardware_requirement": "none", "prover_cooperation": "required", "adversarial_evaluation": "analysis" }, "claims": [], "exposure": { "weights": "partial", "io": "partial", "training": "none", "note": "Recomputation needs the weights and sampled requests inside the checking environment. For closed models, the record describes a trusted, confidential environment; disclosure to the verifier depends on that boundary.", "sources": [ "S-0016", "S-0017", "S-0018" ] }, "family_finding_context": [], "filter_issues": [] } ], "strengths": { "covered": [], "production": [ "M-0001" ], "operationalUse": [ "M-0001" ], "adversarial": [ "M-0024", "M-0001" ], "noNewHardware": [ "M-0023", "M-0010", "M-0001" ], "mitigated": [], "notCounted": [] }, "properties": { "covered": [], "production": [ "M-0001" ], "operationalUse": [ "M-0001" ], "adversarial": [ "M-0024", "M-0001" ], "noNewHardware": [ "M-0023", "M-0010", "M-0001" ], "mitigated": [], "notCounted": [] }, "attack_testing": [ { "id": "M-0023", "record": "M-0023", "evaluation": "analysis", "in_setting": true }, { "id": "M-0024", "record": "M-0024", "evaluation": "independent-red-team", "in_setting": true }, { "id": "M-0010", "record": "M-0010", "evaluation": "red-teamed", "in_setting": true }, { "id": "M-0001", "record": "M-0001", "evaluation": "analysis", "in_setting": true } ], "selected_implementations": {}, "weaknesses": { "gaps": [], "excluded": [], "unlinked": [], "critical": [], "significant": [ { "mech": "M-0023", "n": 4, "historical": false, "title": "Components outside the attested boundary", "classification": "failure", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "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.", "response": null, "sources": [ "S-1500", "S-1501" ] }, { "mech": "M-0023", "n": 5, "historical": false, "title": "Memory-bus interposition extracts attestation keys and forges attestations", "classification": "failure", "kind": "demonstrated-attack", "severity": "significant", "status": "open", "evidence_scope": "inherited", "scope_note": "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.", "related_finding": { "record": "M-0008", "flaw": 1 }, "description": "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.", "response": "Intel and AMD place the physical attack class outside their threat models, according to the researchers. AMD reports firmware fixes for RMPocalypse.", "sources": [ "S-1202", "S-3126", "S-1210", "S-1212", "S-1213", "S-0012", "S-0014", "S-1500", "S-0018" ], "helps": [ { "by": "M-0009", "how": "A tamper-protected enclosure around the chip is the proposed answer when the party that holds the hardware may attack it physically." } ] }, { "mech": "M-0024", "n": 1, "historical": false, "title": "Prompt-controlled entropy inflation widens the covert channel", "classification": "failure", "kind": "demonstrated-attack", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "Gumbel-based inference verification tolerates token choices that honest GPU nondeterminism could produce, and the size of that tolerated set grows with the model's output entropy. Kezins, an independent researcher, showed that an adversary who controls the prompt distribution can raise output entropy and roughly double the bits leaked per token. Across six models of 1 to 32 billion parameters, this cut the slowdown from 146–254 times under benign prompts to 60–118 times. Kezins argues that architectures built on the same unexplained-information bound inherit this attack surface, and recommends calibrating tolerances against local token entropy rather than benign traffic.", "response": null, "sources": [ "S-1507", "S-0015" ], "helps": [ { "by": "M-0002", "how": "Bit-exact replay would remove the tolerance for numerical noise that sets the size of this channel. The record notes that it needs full hardware and software metadata." } ] }, { "mech": "M-0010", "n": 2, "historical": false, "title": "Adversarially disguised fine-tuning partly evades classification", "classification": "failure", "kind": "demonstrated-attack", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "Across 20 evasion strategy families in five monitor-evader rounds, the classifier's accuracy against the most challenging disguised workloads held out from training was 43–87%. White-box LoRA fine-tuning was the only evasion family detected less than half the time. The evaluation covered single nodes, PyTorch-level evasion and NVIDIA hardware.", "response": null, "sources": [ "S-0037" ] }, { "mech": "M-0010", "n": 4, "historical": false, "title": "Counters leak information about protected workloads", "classification": "failure", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "Performance counters have been used as a side channel against TEEs, for example in CounterSEVeillance. NVIDIA disables performance counters in full confidential-computing mode, stating that they could provide an avenue for side-channel attacks. Richer counters for verification therefore pull against confidentiality.", "response": null, "sources": [ "S-0014", "S-1200" ] }, { "mech": "M-0001", "n": 1, "historical": false, "title": "Tolerance for numerical noise leaves a covert channel", "classification": "failure", "kind": "demonstrated-attack", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "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.", "response": null, "sources": [ "S-0020", "S-0015", "S-1507" ], "helps": [ { "by": "M-0002", "how": "Bit-exact inference would remove the numerical tolerance if exact replay can be deployed with the required weights and configuration." } ] }, { "mech": "M-0001", "n": 3, "historical": false, "title": "Some inference optimizations are not covered", "classification": "failure", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "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.", "response": null, "sources": [ "S-1000", "S-0016" ] } ], "criticalMechanisms": [], "significantMechanisms": [ "M-0023", "M-0024", "M-0010", "M-0001" ], "familyContext": [], "scopeLimitations": [ { "mech": "M-0023", "n": 1, "historical": false, "title": "Attestation shows a safeguard ran, not that it is effective", "classification": "scope-limitation", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "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.", "response": null, "sources": [ "S-1500" ] }, { "mech": "M-0023", "n": 2, "historical": false, "title": "Selective attestation leaves traffic uncovered", "classification": "scope-limitation", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "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.", "response": null, "sources": [ "S-1500", "S-0012", "S-0014" ], "helps": [ { "by": "M-0010", "how": "On-chip counters are a proposed route to evidence about everything a chip runs, including a second virtual machine that skips the safeguard." } ] }, { "mech": "M-0023", "n": 3, "historical": false, "title": "Measurements may omit behaviour-relevant configuration or runtime changes", "classification": "scope-limitation", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "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.", "response": null, "sources": [ "S-0014" ] }, { "mech": "M-0024", "n": 2, "historical": false, "title": "Information the declared computation explains is not bounded", "classification": "scope-limitation", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "The bound limits unexplained bits only. Outputs that the declared computation fully explains can still carry valuable information: a compression study notes that an adversary with inference access can extract more proprietary information per bit than naive transmission allows.", "response": null, "sources": [ "S-1508", "S-0019" ] }, { "mech": "M-0024", "n": 3, "historical": false, "title": "Channels other than checked outputs are outside the bound", "classification": "scope-limitation", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "The inference-verification scheme treats side channels as out of scope. A low-trust system design argues that suppressing physical covert bandwidth below kilobits per second is much more achievable than aiming for zero, and that a malicious device can leak one bit of information by deliberately outputting a wrong result.", "response": null, "sources": [ "S-0015", "S-0018" ], "helps": [ { "by": "M-0022", "how": "Physical side channels need separate suppression, which is this mechanism's purpose." } ] }, { "mech": "M-0010", "n": 1, "historical": false, "title": "Software-read telemetry can be forged by the operator", "classification": "scope-limitation", "kind": "theoretical-argument", "severity": "critical", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "NVML-based classification assumes trustworthy telemetry. Without a tamper-resistant read path, an authenticated telemetry channel and secure boot of the monitoring software, an operator who controls the full software stack could forge counter values. Monfared et al. start from the same premise: current GPUs expose little trusted telemetry and can be modified or virtualized.", "response": null, "sources": [ "S-0037", "S-0033" ], "helps": [ { "by": "M-0009", "how": "A guarantee processor on the chip would give the tamper-resistant, authenticated telemetry path the flaw says is missing." } ] }, { "mech": "M-0010", "n": 3, "historical": false, "title": "Timing challenges do not identify the individual chip", "classification": "scope-limitation", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "GEMM and VDF challenges can be answered by identical GPUs elsewhere, and floating-point fingerprints distinguish GPU models, not individual devices. GPU virtualization adds timing leakage that prevents attributing compute use.", "response": null, "sources": [ "S-0033" ] }, { "mech": "M-0001", "n": 2, "historical": false, "title": "Only recorded traffic is checked", "classification": "scope-limitation", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "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.", "response": null, "sources": [ "S-0017", "S-0015" ], "helps": [ { "by": "M-0013", "how": "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." } ] } ], "openQuestions": [ { "mech": "M-0024", "n": 4, "historical": false, "title": "The facility-level design is untested", "classification": "open-question", "kind": "open-question", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "The compute-verification architecture is described with protocol details, potential attacks and prototyping plans, but no prototype results have been published.", "response": null, "sources": [ "S-0019" ] }, { "mech": "M-0010", "n": 5, "historical": false, "title": "No quantified error rates or formal thresholds for timing primitives", "classification": "open-question", "kind": "open-question", "severity": "minor", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "Monfared et al. state that false-positive and false-negative rates are not quantified and leave hardware-specific formal thresholds to future work.", "response": null, "sources": [ "S-0033" ] } ], "minor": 1, "minorFindings": [ { "mech": "M-0001", "n": 4, "historical": false, "title": "Mixed hardware widens the honest baseline", "classification": "failure", "kind": "open-question", "severity": "minor", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "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.", "response": null, "sources": [ "S-0016" ] } ], "minorBy": [ { "id": "M-0001", "n": 1 } ], "notDemonstrated": [], "newChip": [] }, "findings": [ { "mech": "M-0023", "record": "M-0023", "n": 1, "historical": false, "title": "Attestation shows a safeguard ran, not that it is effective", "classification": "scope-limitation", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "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.", "response": null, "sources": [ "S-1500" ] }, { "mech": "M-0023", "record": "M-0023", "n": 2, "historical": false, "title": "Selective attestation leaves traffic uncovered", "classification": "scope-limitation", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "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.", "response": null, "sources": [ "S-1500", "S-0012", "S-0014" ], "helps": [ { "by": "M-0010", "how": "On-chip counters are a proposed route to evidence about everything a chip runs, including a second virtual machine that skips the safeguard." } ] }, { "mech": "M-0023", "record": "M-0023", "n": 3, "historical": false, "title": "Measurements may omit behaviour-relevant configuration or runtime changes", "classification": "scope-limitation", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "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.", "response": null, "sources": [ "S-0014" ] }, { "mech": "M-0023", "record": "M-0023", "n": 4, "historical": false, "title": "Components outside the attested boundary", "classification": "failure", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "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.", "response": null, "sources": [ "S-1500", "S-1501" ] }, { "mech": "M-0023", "record": "M-0023", "n": 5, "historical": false, "title": "Memory-bus interposition extracts attestation keys and forges attestations", "classification": "failure", "kind": "demonstrated-attack", "severity": "significant", "status": "open", "evidence_scope": "inherited", "scope_note": "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.", "related_finding": { "record": "M-0008", "flaw": 1 }, "description": "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.", "response": "Intel and AMD place the physical attack class outside their threat models, according to the researchers. AMD reports firmware fixes for RMPocalypse.", "sources": [ "S-1202", "S-3126", "S-1210", "S-1212", "S-1213", "S-0012", "S-0014", "S-1500", "S-0018" ], "helps": [ { "by": "M-0009", "how": "A tamper-protected enclosure around the chip is the proposed answer when the party that holds the hardware may attack it physically." } ] }, { "mech": "M-0024", "record": "M-0024", "n": 1, "historical": false, "title": "Prompt-controlled entropy inflation widens the covert channel", "classification": "failure", "kind": "demonstrated-attack", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "Gumbel-based inference verification tolerates token choices that honest GPU nondeterminism could produce, and the size of that tolerated set grows with the model's output entropy. Kezins, an independent researcher, showed that an adversary who controls the prompt distribution can raise output entropy and roughly double the bits leaked per token. Across six models of 1 to 32 billion parameters, this cut the slowdown from 146–254 times under benign prompts to 60–118 times. Kezins argues that architectures built on the same unexplained-information bound inherit this attack surface, and recommends calibrating tolerances against local token entropy rather than benign traffic.", "response": null, "sources": [ "S-1507", "S-0015" ], "helps": [ { "by": "M-0002", "how": "Bit-exact replay would remove the tolerance for numerical noise that sets the size of this channel. The record notes that it needs full hardware and software metadata." } ] }, { "mech": "M-0024", "record": "M-0024", "n": 2, "historical": false, "title": "Information the declared computation explains is not bounded", "classification": "scope-limitation", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "The bound limits unexplained bits only. Outputs that the declared computation fully explains can still carry valuable information: a compression study notes that an adversary with inference access can extract more proprietary information per bit than naive transmission allows.", "response": null, "sources": [ "S-1508", "S-0019" ] }, { "mech": "M-0024", "record": "M-0024", "n": 3, "historical": false, "title": "Channels other than checked outputs are outside the bound", "classification": "scope-limitation", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "The inference-verification scheme treats side channels as out of scope. A low-trust system design argues that suppressing physical covert bandwidth below kilobits per second is much more achievable than aiming for zero, and that a malicious device can leak one bit of information by deliberately outputting a wrong result.", "response": null, "sources": [ "S-0015", "S-0018" ], "helps": [ { "by": "M-0022", "how": "Physical side channels need separate suppression, which is this mechanism's purpose." } ] }, { "mech": "M-0024", "record": "M-0024", "n": 4, "historical": false, "title": "The facility-level design is untested", "classification": "open-question", "kind": "open-question", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "The compute-verification architecture is described with protocol details, potential attacks and prototyping plans, but no prototype results have been published.", "response": null, "sources": [ "S-0019" ] }, { "mech": "M-0010", "record": "M-0010", "n": 1, "historical": false, "title": "Software-read telemetry can be forged by the operator", "classification": "scope-limitation", "kind": "theoretical-argument", "severity": "critical", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "NVML-based classification assumes trustworthy telemetry. Without a tamper-resistant read path, an authenticated telemetry channel and secure boot of the monitoring software, an operator who controls the full software stack could forge counter values. Monfared et al. start from the same premise: current GPUs expose little trusted telemetry and can be modified or virtualized.", "response": null, "sources": [ "S-0037", "S-0033" ], "helps": [ { "by": "M-0009", "how": "A guarantee processor on the chip would give the tamper-resistant, authenticated telemetry path the flaw says is missing." } ] }, { "mech": "M-0010", "record": "M-0010", "n": 2, "historical": false, "title": "Adversarially disguised fine-tuning partly evades classification", "classification": "failure", "kind": "demonstrated-attack", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "Across 20 evasion strategy families in five monitor-evader rounds, the classifier's accuracy against the most challenging disguised workloads held out from training was 43–87%. White-box LoRA fine-tuning was the only evasion family detected less than half the time. The evaluation covered single nodes, PyTorch-level evasion and NVIDIA hardware.", "response": null, "sources": [ "S-0037" ] }, { "mech": "M-0010", "record": "M-0010", "n": 3, "historical": false, "title": "Timing challenges do not identify the individual chip", "classification": "scope-limitation", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "GEMM and VDF challenges can be answered by identical GPUs elsewhere, and floating-point fingerprints distinguish GPU models, not individual devices. GPU virtualization adds timing leakage that prevents attributing compute use.", "response": null, "sources": [ "S-0033" ] }, { "mech": "M-0010", "record": "M-0010", "n": 4, "historical": false, "title": "Counters leak information about protected workloads", "classification": "failure", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "Performance counters have been used as a side channel against TEEs, for example in CounterSEVeillance. NVIDIA disables performance counters in full confidential-computing mode, stating that they could provide an avenue for side-channel attacks. Richer counters for verification therefore pull against confidentiality.", "response": null, "sources": [ "S-0014", "S-1200" ] }, { "mech": "M-0010", "record": "M-0010", "n": 5, "historical": false, "title": "No quantified error rates or formal thresholds for timing primitives", "classification": "open-question", "kind": "open-question", "severity": "minor", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "Monfared et al. state that false-positive and false-negative rates are not quantified and leave hardware-specific formal thresholds to future work.", "response": null, "sources": [ "S-0033" ] }, { "mech": "M-0001", "record": "M-0001", "n": 1, "historical": false, "title": "Tolerance for numerical noise leaves a covert channel", "classification": "failure", "kind": "demonstrated-attack", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "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.", "response": null, "sources": [ "S-0020", "S-0015", "S-1507" ], "helps": [ { "by": "M-0002", "how": "Bit-exact inference would remove the numerical tolerance if exact replay can be deployed with the required weights and configuration." } ] }, { "mech": "M-0001", "record": "M-0001", "n": 2, "historical": false, "title": "Only recorded traffic is checked", "classification": "scope-limitation", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "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.", "response": null, "sources": [ "S-0017", "S-0015" ], "helps": [ { "by": "M-0013", "how": "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." } ] }, { "mech": "M-0001", "record": "M-0001", "n": 3, "historical": false, "title": "Some inference optimizations are not covered", "classification": "failure", "kind": "theoretical-argument", "severity": "significant", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "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.", "response": null, "sources": [ "S-1000", "S-0016" ] }, { "mech": "M-0001", "record": "M-0001", "n": 4, "historical": false, "title": "Mixed hardware widens the honest baseline", "classification": "failure", "kind": "open-question", "severity": "minor", "status": "open", "evidence_scope": null, "scope_note": null, "related_finding": null, "description": "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.", "response": null, "sources": [ "S-0016" ] } ], "possible_additions": [ { "id": "M-0002", "title": "Deterministic and bit-exact inference", "url": "https://trustbutveri.fyi/mechanisms/deterministic-inference/", "readiness": "R3", "development_status": { "code": "R3", "label": "Operational use", "short": "Operational use", "rank": 3, "legacy_code": "R3" }, "security_evidence": { "attack_testing": { "status": "analysis", "label": "Published security analysis", "kind": "analysis" }, "independent_evaluation": { "status": "unassessed" }, "formal_proof": { "status": "unassessed" }, "deployment_assurance": { "status": "unassessed" }, "legacy_evaluation_code": null, "scoped_findings": [], "open_failures": { "critical": 0, "significant": 0, "minor": 0 } }, "fits_filters": true, "filter_issues": [], "reasons": [ { "kind": "flaw", "mech": "M-0024", "n": 1, "title": "Prompt-controlled entropy inflation widens the covert channel", "severity": "significant", "how": "Bit-exact replay would remove the tolerance for numerical noise that sets the size of this channel. The record notes that it needs full hardware and software metadata." }, { "kind": "flaw", "mech": "M-0001", "n": 1, "title": "Tolerance for numerical noise leaves a covert channel", "severity": "significant", "how": "Bit-exact inference would remove the numerical tolerance if exact replay can be deployed with the required weights and configuration." }, { "kind": "blocker", "mech": "M-0024", "text": "Tolerance for numerical nondeterminism sets the size of the residual channel; bit-exact replay would remove it but needs full hardware and software metadata." } ] }, { "id": "M-0009", "title": "Hardware-enabled guarantees (flexHEG) and guarantee processors", "url": "https://trustbutveri.fyi/mechanisms/flexheg-guarantee-processors/", "readiness": "R1", "development_status": { "code": "R1", "label": "Proposed", "short": "Proposed", "rank": 1, "legacy_code": "R1" }, "security_evidence": { "attack_testing": { "status": "analysis", "label": "Published security analysis", "kind": "analysis" }, "independent_evaluation": { "status": "unassessed" }, "formal_proof": { "status": "unassessed" }, "deployment_assurance": { "status": "unassessed" }, "legacy_evaluation_code": null, "scoped_findings": [ { "n": 1, "severity": "significant", "status": "open", "evidence_scope": "unassessed", "sources": [ "S-1204", "S-0057" ] }, { "n": 2, "severity": "significant", "status": "open", "evidence_scope": "unassessed", "sources": [ "S-1204" ] }, { "n": 4, "severity": "significant", "status": "open", "evidence_scope": "unassessed", "sources": [ "S-1204" ] } ], "open_failures": { "critical": 0, "significant": 3, "minor": 0 } }, "fits_filters": true, "filter_issues": [], "reasons": [ { "kind": "flaw", "mech": "M-0023", "n": 5, "title": "Memory-bus interposition extracts attestation keys and forges attestations", "severity": "significant", "how": "A tamper-protected enclosure around the chip is the proposed answer when the party that holds the hardware may attack it physically." }, { "kind": "blocker", "mech": "M-0010", "text": "Shipping accelerators need a tamper-resistant, authenticated telemetry path." } ] }, { "id": "M-0012", "title": "Hardware-attested weight binding", "url": "https://trustbutveri.fyi/mechanisms/model-identity-attestation/", "readiness": "R3", "development_status": { "code": "R3", "label": "Operational use", "short": "Operational use", "rank": 3, "legacy_code": "R3" }, "security_evidence": { "attack_testing": { "status": "independent-red-team", "label": "Published attack testing", "kind": "practical", "attribution": "Independent team" }, "independent_evaluation": { "status": "unassessed" }, "formal_proof": { "status": "unassessed" }, "deployment_assurance": { "status": "unassessed" }, "legacy_evaluation_code": null, "scoped_findings": [ { "n": 1, "severity": "critical", "status": "open", "evidence_scope": "inherited", "sources": [ "S-1202", "S-1210", "S-1212", "S-1213", "S-1206", "S-0012" ], "related_finding": { "record": "M-0008", "flaw": 1 } }, { "n": 2, "severity": "significant", "status": "mitigated", "evidence_scope": "unassessed", "sources": [ "S-0013", "S-0014" ] } ], "open_failures": { "critical": 1, "significant": 0, "minor": 0 } }, "fits_filters": true, "filter_issues": [], "reasons": [ { "kind": "blocker", "mech": "M-0023", "text": "Safeguard evidence must be bound to the model actually served, which depends on model-identity attestation." } ] }, { "id": "M-0008", "title": "TEE remote attestation for AI workloads", "url": "https://trustbutveri.fyi/mechanisms/tee-remote-attestation/", "readiness": "R3", "development_status": { "code": "R3", "label": "Operational use", "short": "Operational use", "rank": 3, "legacy_code": "R3" }, "security_evidence": { "attack_testing": { "status": "independent-red-team", "label": "Published attack testing", "kind": "practical", "attribution": "Independent team" }, "independent_evaluation": { "status": "unassessed" }, "formal_proof": { "status": "unassessed" }, "deployment_assurance": { "status": "unassessed" }, "legacy_evaluation_code": null, "scoped_findings": [ { "n": 1, "severity": "critical", "status": "open", "evidence_scope": "mechanism", "sources": [ "S-1202", "S-3126", "S-0012", "S-1206", "S-0014", "S-1210", "S-1212", "S-3127" ] }, { "n": 2, "severity": "critical", "status": "open", "evidence_scope": "mechanism", "sources": [ "S-1210", "S-1211" ] }, { "n": 3, "severity": "critical", "status": "mitigated", "evidence_scope": "mechanism", "sources": [ "S-1212", "S-1213", "S-3127", "S-3128" ] }, { "n": 4, "severity": "significant", "status": "open", "evidence_scope": "mechanism", "sources": [ "S-1202" ] }, { "n": 5, "severity": "significant", "status": "open", "evidence_scope": "mechanism", "sources": [ "S-0012", "S-0009", "S-0014", "S-3123", "S-3129" ] }, { "n": 8, "severity": "significant", "status": "open", "evidence_scope": "mechanism", "sources": [ "S-0014", "S-0009", "S-3130", "S-3131" ] } ], "open_failures": { "critical": 2, "significant": 3, "minor": 0 } }, "fits_filters": true, "filter_issues": [], "reasons": [ { "kind": "blocker", "mech": "M-0023", "text": "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." }, { "kind": "blocker", "mech": "M-0023", "text": "Trust rests on a small number of hardware vendors, and a per-CPU Intel attestation key has been extracted by physical attack." }, { "kind": "prerequisite", "mech": "M-0010" } ] }, { "id": "M-0014", "title": "Bandwidth limits and compartmentalization", "url": "https://trustbutveri.fyi/mechanisms/bandwidth-limits-and-compartmentalization/", "readiness": "R2", "development_status": { "code": "R2", "label": "Research demonstration", "short": "Research demo", "rank": 2, "legacy_code": "R2" }, "security_evidence": { "attack_testing": { "status": "analysis", "label": "Published security analysis", "kind": "analysis" }, "independent_evaluation": { "status": "unassessed" }, "formal_proof": { "status": "unassessed" }, "deployment_assurance": { "status": "unassessed" }, "legacy_evaluation_code": null, "scoped_findings": [ { "n": 2, "severity": "significant", "status": "open", "evidence_scope": "unassessed", "sources": [ "S-1301" ] }, { "n": 5, "severity": "significant", "status": "open", "evidence_scope": "unassessed", "sources": [ "S-1313" ] } ], "open_failures": { "critical": 0, "significant": 2, "minor": 0 } }, "fits_filters": true, "filter_issues": [], "reasons": [ { "kind": "blocker", "mech": "M-0024", "text": "The prover's compute must be isolated so that all traffic passes through the verifier's interlock; any unmonitored path voids the bound." } ] }, { "id": "M-0013", "title": "Network taps and certifiers", "url": "https://trustbutveri.fyi/mechanisms/network-taps-and-certifiers/", "readiness": "R1", "development_status": { "code": "R1", "label": "Proposed", "short": "Proposed", "rank": 1, "legacy_code": "R1" }, "security_evidence": { "attack_testing": { "status": "analysis", "label": "Published security analysis", "kind": "analysis" }, "independent_evaluation": { "status": "unassessed" }, "formal_proof": { "status": "unassessed" }, "deployment_assurance": { "status": "unassessed" }, "legacy_evaluation_code": null, "scoped_findings": [ { "n": 1, "severity": "significant", "status": "open", "evidence_scope": "unassessed", "sources": [ "S-1300", "S-1507" ] }, { "n": 4, "severity": "significant", "status": "open", "evidence_scope": "unassessed", "sources": [ "S-1312" ] }, { "n": 6, "severity": "minor", "status": "mitigated", "evidence_scope": "unassessed", "sources": [ "S-1300" ] } ], "open_failures": { "critical": 0, "significant": 2, "minor": 0 } }, "fits_filters": true, "filter_issues": [], "reasons": [ { "kind": "blocker", "mech": "M-0001", "text": "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." }, { "kind": "prerequisite", "mech": "M-0024" } ] }, { "id": "M-0022", "title": "Side-channel suppression for isolated facilities", "url": "https://trustbutveri.fyi/mechanisms/side-channel-suppression/", "readiness": "R1", "development_status": { "code": "R1", "label": "Proposed", "short": "Proposed", "rank": 1, "legacy_code": "R1" }, "security_evidence": { "attack_testing": { "status": "analysis", "label": "Published security analysis", "kind": "analysis" }, "independent_evaluation": { "status": "unassessed" }, "formal_proof": { "status": "unassessed" }, "deployment_assurance": { "status": "unassessed" }, "legacy_evaluation_code": null, "scoped_findings": [], "open_failures": { "critical": 0, "significant": 0, "minor": 0 } }, "fits_filters": true, "filter_issues": [], "reasons": [ { "kind": "blocker", "mech": "M-0024", "text": "Physical side channels need separate suppression, and one design treats a low residual bandwidth, rather than zero, as the realistic target." } ] } ], "goal": null, "design": null, "dependencies": { "prerequisites": [ { "id": "M-0008", "neededBy": [ "M-0023", "M-0010" ] }, { "id": "M-0012", "neededBy": [ "M-0023" ] }, { "id": "M-0014", "neededBy": [ "M-0024" ] }, { "id": "M-0022", "neededBy": [ "M-0024" ] }, { "id": "M-0013", "neededBy": [ "M-0024" ] } ], "shared": [ { "id": "M-0008", "by": [ "M-0023", "M-0010" ], "inProposal": false }, { "id": "M-0013", "by": [ "M-0024", "M-0001" ], "inProposal": false } ], "blockers": [ { "mech": "M-0023", "n": 1, "historical": false, "text": "No published design shows that all of a provider's traffic passes through the attested safeguard path; current evidence covers individual attested responses.", "theme": "coverage-hidden-compute", "blocked_by": null, "sources": [ "S-1500", "S-0014" ], "inProposal": null }, { "mech": "M-0023", "n": 2, "historical": false, "text": "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.", "theme": "performance-compatibility", "blocked_by": "M-0008", "sources": [ "S-0014", "S-0009" ], "inProposal": false }, { "mech": "M-0023", "n": 3, "historical": false, "text": "Trust rests on a small number of hardware vendors, and a per-CPU Intel attestation key has been extracted by physical attack.", "theme": "hardware-trust", "blocked_by": "M-0008", "sources": [ "S-0014", "S-1202" ], "inProposal": false }, { "mech": "M-0023", "n": 4, "historical": false, "text": "Safeguard evidence must be bound to the model actually served, which depends on model-identity attestation.", "theme": "evidence-binding", "blocked_by": "M-0012", "sources": [ "S-0009", "S-0013" ], "inProposal": false }, { "mech": "M-0023", "n": 5, "historical": false, "text": "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.", "theme": "adversarial-validation", "blocked_by": null, "sources": [ "S-1501", "S-1504" ], "inProposal": null }, { "mech": "M-0024", "n": 1, "historical": false, "text": "The prover's compute must be isolated so that all traffic passes through the verifier's interlock; any unmonitored path voids the bound.", "theme": "coverage-hidden-compute", "blocked_by": "M-0014", "sources": [ "S-0019" ], "inProposal": false }, { "mech": "M-0024", "n": 2, "historical": false, "text": "Physical side channels need separate suppression, and one design treats a low residual bandwidth, rather than zero, as the realistic target.", "theme": "coverage-hidden-compute", "blocked_by": "M-0022", "sources": [ "S-0018", "S-0015" ], "inProposal": false }, { "mech": "M-0024", "n": 3, "historical": false, "text": "Tolerance for numerical nondeterminism sets the size of the residual channel; bit-exact replay would remove it but needs full hardware and software metadata.", "theme": "protocol-soundness", "blocked_by": "M-0002", "sources": [ "S-1507", "S-0018" ], "inProposal": false }, { "mech": "M-0024", "n": 4, "historical": false, "text": "Recomputation over confidential weights and inputs needs a protected setting: prover recomputation in a verifier-controlled enclosure, verifier recomputation in a prover-controlled enclosure, or zero-knowledge proofs.", "theme": "privacy-leakage", "blocked_by": null, "sources": [ "S-0019" ], "inProposal": null }, { "mech": "M-0024", "n": 5, "historical": false, "text": "No prototype of the facility-level architecture exists to red-team.", "theme": "adversarial-validation", "blocked_by": null, "sources": [ "S-0019" ], "inProposal": null }, { "mech": "M-0010", "n": 1, "historical": false, "text": "Shipping accelerators need a tamper-resistant, authenticated telemetry path.", "theme": "hardware-trust", "blocked_by": "M-0009", "sources": [ "S-0037", "S-0034" ], "inProposal": false }, { "mech": "M-0010", "n": 2, "historical": false, "text": "NVIDIA's full confidential-computing mode disables the hardware performance counters its profiling tools use, so telemetry that needs them conflicts with it.", "theme": "privacy-leakage", "blocked_by": null, "sources": [ "S-1200", "S-0014" ], "inProposal": null }, { "mech": "M-0010", "n": 3, "historical": false, "text": "Continuous challenge puzzles cost power and throughput on production workloads.", "theme": "performance-compatibility", "blocked_by": null, "sources": [ "S-0033" ], "inProposal": null }, { "mech": "M-0010", "n": 4, "historical": false, "text": "Evaluation has not gone beyond single nodes, framework-level evasion and one vendor's hardware.", "theme": "adversarial-validation", "blocked_by": null, "sources": [ "S-0037" ], "inProposal": null }, { "mech": "M-0001", "n": 1, "historical": false, "text": "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.", "theme": "coverage-hidden-compute", "blocked_by": "M-0013", "sources": [ "S-1008", "S-0067" ], "inProposal": false }, { "mech": "M-0001", "n": 2, "historical": false, "text": "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'.", "theme": "hardware-trust", "blocked_by": null, "sources": [ "S-1008", "S-0015" ], "inProposal": null }, { "mech": "M-0001", "n": 3, "historical": false, "text": "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'.", "theme": "adversarial-validation", "blocked_by": null, "sources": [ "S-1008", "S-1507" ], "inProposal": null }, { "mech": "M-0001", "n": 4, "historical": false, "text": "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.", "theme": "performance-compatibility", "blocked_by": null, "sources": [ "S-0016", "S-1006" ], "inProposal": null }, { "mech": "M-0001", "n": 5, "historical": false, "text": "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.", "theme": "privacy-leakage", "blocked_by": null, "sources": [ "S-0016", "S-0017", "S-0018" ], "inProposal": null } ] }, "exposure": { "weights": { "shown": [], "partial": [ "M-0023", "M-0024", "M-0010", "M-0001" ], "hidden": [], "none": [], "unknown": [] }, "io": { "shown": [], "partial": [ "M-0023", "M-0024", "M-0010", "M-0001" ], "hidden": [], "none": [], "unknown": [] }, "training": { "shown": [], "partial": [ "M-0010" ], "hidden": [], "none": [ "M-0023", "M-0024", "M-0001" ], "unknown": [] } }, "implementations": [ { "mechanism": "M-0023", "selected": null, "implementations": [] }, { "mechanism": "M-0024", "selected": null, "implementations": [] }, { "mechanism": "M-0010", "selected": null, "implementations": [] }, { "mechanism": "M-0001", "selected": null, "implementations": [ { "id": "I-0011", "title": "AI 2040 inference-only verification stack", "url": "https://trustbutveri.fyi/implementations/ai-2040-inference-only-verification-plan/" }, { "id": "I-0002", "title": "DiFR (Divergence From Reference)", "url": "https://trustbutveri.fyi/implementations/difr/" }, { "id": "I-0012", "title": "Low-trust AI compute verification system overview", "url": "https://trustbutveri.fyi/implementations/low-trust-compute-verification-system-overview/" }, { "id": "I-0008", "title": "SASH confidential network logger", "url": "https://trustbutveri.fyi/implementations/sash-confidential-network-logger/" }, { "id": "I-0001", "title": "TOPLOC", "url": "https://trustbutveri.fyi/implementations/toploc/" } ] } ], "sources": [ { "id": "S-1500", "title": "Proof-of-Guardrail in AI Agents and What (Not) to Trust from It", "authors": "X. Jin et al.", "year": 2026, "url": "https://arxiv.org/abs/2603.05786", "path": "/sources/jin-proof-of-guardrail/" }, { "id": "S-1501", "title": "Verifiable-ClawGuard: proof-of-guardrail reference code", "authors": "SaharaLabsAI", "year": 2026, "url": "https://github.com/SaharaLabsAI/Verifiable-ClawGuard", "path": "/sources/sahara-verifiable-clawguard-code/" }, { "id": "S-3362", "title": "Safety Without Compromising on Privacy", "authors": "D. McCann-Sayles et al.", "year": 2026, "url": "https://tinfoil.sh/blog/2026-09-14-safety-without-compromising-privacy", "path": "/sources/tinfoil-safety-without-compromising-privacy/" }, { "id": "S-0012", "title": "PAL*M: Property Attestation for Large Generative Models", "authors": "P. Chantasantitam et al.", "year": 2026, "url": "https://arxiv.org/abs/2601.16199", "path": "/sources/chantasantitam-palm/" }, { "id": "S-1503", "title": "Enabling Verifiably-Scoped Monitoring through Large Language Models and Trusted Compute", "authors": "B. Penchas et al.", "year": 2026, "url": "https://icml.cc/virtual/2026/78630", "path": "/sources/penchas-verifiably-scoped-monitoring/" }, { "id": "S-1504", "title": "Auditor-in-a-Box: Tools for Third-Party Auditing", "authors": "R. Rinberg & B. Penchas", "year": 2026, "url": "https://www.lesswrong.com/posts/uWYk7MM9hAf9GEbGe/auditor-in-a-box-tools-for-third-party-auditing", "path": "/sources/rinberg-auditor-in-a-box/" }, { "id": "S-1202", "title": "TEE.fail: Breaking Trusted Execution Environments via DDR5 Memory Bus Interposition", "authors": "J. Chuang et al.", "year": 2026, "url": "https://tee.fail/", "path": "/sources/chuang-tee-fail/" }, { "id": "S-3126", "title": "DDRop: Active Memory Interposer Attacks on Confidential VMs by Dropping DDR5 Writes", "authors": "J. De Meulemeester et al.", "year": 2026, "url": "https://ddropattack.eu/", "path": "/sources/de-meulemeester-ddrop/" }, { "id": "S-0014", "title": "On TEEs for Privacy-Preserving Monitoring in AI Governance", "authors": "Gloria Z", "year": 2026, "url": "https://techgov.intelligence.org/blog/on-tees-for-privacy-preserving-monitoring-in-ai-governance", "path": "/sources/zhao-tees-privacy-preserving-monitoring/" }, { "id": "S-1210", "title": "Battering RAM: Low-Cost Interposer Attacks on Confidential Computing via Dynamic Memory Aliasing", "authors": "J. De Meulemeester et al.", "year": 2026, "url": "https://batteringram.eu/", "path": "/sources/de-meulemeester-battering-ram/" }, { "id": "S-1212", "title": "RMPocalypse: How a Catch-22 Breaks AMD SEV-SNP", "authors": "B. Schlüter & S. Shinde", "year": 2025, "url": "https://rmpocalypse.github.io/", "path": "/sources/schluter-rmpocalypse/" }, { "id": "S-1213", "title": "SEV-SNP RMP Initialization Vulnerability (AMD-SB-3020)", "authors": "AMD", "year": 2025, "url": "https://www.amd.com/en/resources/product-security/bulletin/amd-sb-3020.html", "path": "/sources/amd-sb-3020-rmp-initialization/" }, { "id": "S-0018", "title": "A System Overview for Near-Term, Low-Trust AI Compute Verification", "authors": "N. Cankaya", "year": 2026, "url": "https://intelligence.org/wp-content/uploads/2026/06/A-system-overview-for-near-term-low-trust-AI-compute-verification.pdf", "path": "/sources/cankaya-system-overview-low-trust-compute-verification/" }, { "id": "S-0019", "title": "Verifying AI Compute by Bounding Unexplained Information Exfiltration", "authors": "J. Petrie & Y. Mühlhäuser", "year": 2026, "url": "https://openreview.net/forum?id=qtgG5HZSsk", "path": "/sources/petrie-bounding-unexplained-information-exfiltration/" }, { "id": "S-0015", "title": "Verifying LLM Inference to Detect Model Weight Exfiltration", "authors": "R. Rinberg et al.", "year": 2025, "url": "https://arxiv.org/abs/2511.02620", "path": "/sources/rinberg-verifying-llm-inference-weight-exfiltration/" }, { "id": "S-1507", "title": "Adversarial Entropy Inflation Against Gumbel-Based Inference Verification", "authors": "N. Kezins", "year": 2026, "url": "https://arxiv.org/abs/2608.23375", "path": "/sources/kezins-adversarial-entropy-inflation/" }, { "id": "S-1508", "title": "Haiku to Opus in Just 10 bits: LLMs Unlock Large Compression Gains", "authors": "R. Rinberg et al.", "year": 2026, "url": "https://arxiv.org/abs/2604.02343", "path": "/sources/rinberg-haiku-to-opus-compression/" }, { "id": "S-0033", "title": "Timing and Memory Telemetry on GPUs for AI Governance", "authors": "S. K. Monfared et al.", "year": 2026, "url": "https://arxiv.org/abs/2602.09369", "path": "/sources/monfared-timing-memory-telemetry-gpus/" }, { "id": "S-0034", "title": "Guaranteeable Memory: An HBM-Based Chiplet for Verifiable AI Workloads", "authors": "J. Petrie", "year": 2025, "url": "https://openreview.net/forum?id=uc79kOv0MV", "path": "/sources/petrie-guaranteeable-memory/" }, { "id": "S-0037", "title": "Detecting Hidden ML Training With Zero-Overhead Telemetry", "authors": "R. Rahman & S. Tajdari", "year": 2026, "url": "https://arxiv.org/abs/2606.19262", "path": "/sources/rahman-detecting-hidden-ml-training/" }, { "id": "S-1200", "title": "NVIDIA Secure AI with Blackwell and Hopper GPUs (White Paper)", "authors": "NVIDIA", "year": 2025, "url": "https://docs.nvidia.com/nvidia-secure-ai-with-blackwell-and-hopper-gpus-whitepaper.pdf", "path": "/sources/nvidia-secure-ai-blackwell-hopper-whitepaper/" }, { "id": "S-0016", "title": "DiFR: Inference Verification Despite Nondeterminism", "authors": "A. Karvonen et al.", "year": 2025, "url": "https://arxiv.org/abs/2511.20621", "path": "/sources/karvonen-difr/" }, { "id": "S-1000", "title": "TOPLOC: A Locality Sensitive Hashing Scheme for Trustless Verifiable Inference", "authors": "J. M. Ong et al.", "year": 2025, "url": "https://proceedings.mlr.press/v267/ong25a.html", "path": "/sources/ong-toploc/" }, { "id": "S-1001", "title": "PrimeIntellect-ai/toploc (GitHub repository)", "authors": "Prime Intellect", "year": 2025, "url": "https://github.com/PrimeIntellect-ai/toploc", "path": "/sources/primeintellect-toploc-code/" }, { "id": "S-1003", "title": "INTELLECT-2: A Reasoning Model Trained Through Globally Decentralized Reinforcement Learning", "authors": "Prime Intellect Team et al.", "year": 2025, "url": "https://arxiv.org/abs/2505.07291", "path": "/sources/primeintellect-intellect-2/" }, { "id": "S-1005", "title": "adamkarvonen/difr (GitHub repository)", "authors": "A. Karvonen", "year": 2025, "url": "https://github.com/adamkarvonen/difr", "path": "/sources/karvonen-difr-code/" }, { "id": "S-1006", "title": "Scaling Recomputation Inference Verification", "authors": "Amodo Design", "year": 2026, "url": "https://amododesign.com/notes/2026-09-02-scaling-recomputation-inference-verification/", "path": "/sources/amodo-scaling-recomputation-inference-verification/" }, { "id": "S-1008", "title": "AI 2040 Plan A — Verification SITREP", "authors": "Amodo Design", "year": 2026, "url": "https://amododesign.com/ai-verification/plan-a-sitrep/", "path": "/sources/amodo-plan-a-verification-sitrep/" }, { "id": "S-3000", "title": "SYNTHETIC-2 Release: Four Million Collaboratively Generated Reasoning Traces", "authors": "Prime Intellect", "year": 2025, "url": "https://www.primeintellect.ai/blog/synthetic-2-release", "path": "/sources/primeintellect-synthetic-2-release/" }, { "id": "S-3001", "title": "An Inference Verification Prototype — Stage 1", "authors": "Amodo Design", "year": 2026, "url": "https://amododesign.com/notes/2026-06-29-inference-verification-prototype/", "path": "/sources/amodo-inference-verification-prototype-stage-1/" }, { "id": "S-0020", "title": "Bit-Exact AI Inference Verification Without Performance Tradeoffs", "authors": "N. Cankaya", "year": 2026, "url": "https://arxiv.org/abs/2606.00279", "path": "/sources/cankaya-bit-exact-inference-verification/" }, { "id": "S-0017", "title": "Example Schemes for Verifying High-Stakes AI Agreements", "authors": "Amodo Design", "year": 2026, "url": "https://amododesign.com/notes/2026-06-23-verification-algorithms/", "path": "/sources/amodo-example-schemes-high-stakes-ai-agreements/" }, { "id": "S-0009", "title": "Attestable Audits: Verifiable AI Safety Benchmarks Using Trusted Execution Environments", "authors": "C. Schnabl et al.", "year": 2025, "url": "https://arxiv.org/abs/2506.23706", "path": "/sources/schnabl-attestable-audits/" }, { "id": "S-0013", "title": "How Tinfoil Proves Exactly What Model Is Running", "authors": "Tinfoil Team", "year": 2026, "url": "https://tinfoil.sh/blog/2026-02-03-proving-model-identity", "path": "/sources/tinfoil-proving-model-identity/" }, { "id": "S-0067", "title": "Verification Plan", "authors": "R. Dean", "year": 2026, "url": "https://ai-2040.com/supplements/verification-plan", "path": "/sources/dean-verification-plan/" } ] }