{
  "format": "agent-article-v0",
  "article_id": "EA-753",
  "canonical_url": "/tools/ea-753-confidential-ai-inference-remote-attestation",
  "page_type": "reference",
  "claim_ids": [],
  "evidence_ids": [
    "EVD-EA-753-URL-375497BF530B",
    "EVD-EA-753-URL-545EF7FD8283",
    "EVD-EA-753-URL-66DED0CA69DC",
    "EVD-EA-753-URL-D651AC3D8662",
    "EVD-EA-753-URL-BC7B5197D662",
    "EA-753-CURRENT-APPROVAL",
    "EA-753-CURRENT-BODY-READBACK"
  ],
  "unknown_ids": [
    "EA-753-UNKNOWN-01"
  ],
  "operational_object_ids": [],
  "provider": [],
  "models": [],
  "plans": [],
  "surfaces": [],
  "regions": [],
  "current_status": "CURRENT",
  "observation_status": "SPEC_ONLY",
  "freshness": "CURRENT",
  "last_verified": "2026-10-20",
  "supersedes": [],
  "superseded_by": [],
  "provenance": [
    "EA753-CANDIDATE-20261025-ORPHAN-RESCUE-C5",
    "SRC-EA-753-URL-375497BF530B",
    "SRC-EA-753-URL-545EF7FD8283",
    "SRC-EA-753-URL-66DED0CA69DC",
    "SRC-EA-753-URL-D651AC3D8662",
    "SRC-EA-753-URL-BC7B5197D662",
    "BODY-SHA256:77e1370f49495ab13f80f43395eb1dc768bc9c6ee5320016b7b7eb3b9d7143da"
  ],
  "article_sources": [
    {
      "id": "EVD-EA-753-URL-375497BF530B",
      "evidence_class": "RESEARCH",
      "title": "Google Cloud, Confidential VM attestation:",
      "href": "https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/attestation",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-753-URL-375497BF530B"
      ],
      "supporting_locators": [
        "Source / provenance locators",
        "- Google Cloud, Confidential VM attestation: https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/attestation"
      ],
      "scope": "Source / provenance locators",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-753-URL-545EF7FD8283",
      "evidence_class": "RESEARCH",
      "title": "Google Cloud, Confidential VM overview:",
      "href": "https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/confidential-vm-overview",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-753-URL-545EF7FD8283"
      ],
      "supporting_locators": [
        "Source / provenance locators",
        "- Google Cloud, Confidential VM overview: https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/confidential-vm-overview"
      ],
      "scope": "Source / provenance locators",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-753-URL-66DED0CA69DC",
      "evidence_class": "RESEARCH",
      "title": "NVIDIA Attestation SDK:",
      "href": "https://docs.nvidia.com/attestation/attestation-client-tools-sdk/latest/gpu_and_switch_attestation.html",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-753-URL-66DED0CA69DC"
      ],
      "supporting_locators": [
        "Source / provenance locators",
        "- NVIDIA Attestation SDK: https://docs.nvidia.com/attestation/attestation-client-tools-sdk/latest/gpu_and_switch_attestation.html"
      ],
      "scope": "Source / provenance locators",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-753-URL-D651AC3D8662",
      "evidence_class": "RESEARCH",
      "title": "NVIDIA Confidential Containers Attestation:",
      "href": "https://docs.nvidia.com/datacenter/cloud-native/confidential-containers/latest/attestation.html",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-753-URL-D651AC3D8662"
      ],
      "supporting_locators": [
        "Source / provenance locators",
        "- NVIDIA Confidential Containers Attestation: https://docs.nvidia.com/datacenter/cloud-native/confidential-containers/latest/attestation.html"
      ],
      "scope": "Source / provenance locators",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-753-URL-BC7B5197D662",
      "evidence_class": "RESEARCH",
      "title": "NVIDIA Hopper Remote Attestation Example:",
      "href": "https://docs.nvidia.com/attestation/quick-start-guide/latest/attestation-examples/hopper_single_gpu.html",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-753-URL-BC7B5197D662"
      ],
      "supporting_locators": [
        "Source / provenance locators",
        "- NVIDIA Hopper Remote Attestation Example: https://docs.nvidia.com/attestation/quick-start-guide/latest/attestation-examples/hopper_single_gpu.html"
      ],
      "scope": "Source / provenance locators",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EA-753-CURRENT-APPROVAL",
      "evidence_class": "RESEARCH",
      "title": "現行本文の監査・revision対応",
      "supports": [
        "現行台帳のAUDIT_PASSと候補・revision・本文ハッシュの対応を確認しています。"
      ],
      "limit": "監査状態と来歴の記録です。外部根拠の最新性や本文の全主張を追加保証しません。",
      "origin_class": "AUDITED_CANONICAL_RESEARCH_PACKET",
      "source_or_experiment_ids": [
        "EA753-CANDIDATE-20261025-ORPHAN-RESCUE-C5"
      ],
      "supporting_locators": [
        "1KqhJtiSb2S84rWmeYVwKcT_76dBwZ_B0ZZuAWQqH0rQ/EA_REVISION_LEDGER#ea_id=EA-753&revision_id=EA-753-RESEARCH-20261020-C4-R1&candidate_id=EA753-CANDIDATE-20261025-ORPHAN-RESCUE-C5",
        "EA-753-RESEARCH-20261020-C4-R1"
      ],
      "provenance_status": "CURRENT_CANONICAL_APPROVAL_READBACK"
    },
    {
      "id": "EA-753-CURRENT-BODY-READBACK",
      "evidence_class": "IMPL",
      "title": "現行本文の収録照合",
      "supports": [
        "公開生成入力の本文SHA-256が、監査済み候補に結び付いたハッシュと一致しています。"
      ],
      "limit": "本文の同一性の検査です。外部システムの挙動や全主張の真偽を実証したものではありません。",
      "checked_at": "2026-10-25",
      "origin_class": "IMPLEMENTATION_READBACK",
      "source_or_experiment_ids": [
        "BODY-SHA256:77e1370f49495ab13f80f43395eb1dc768bc9c6ee5320016b7b7eb3b9d7143da"
      ],
      "supporting_locators": [
        "content/staging/EA-753.md",
        "77e1370f49495ab13f80f43395eb1dc768bc9c6ee5320016b7b7eb3b9d7143da",
        "EA-753-RESEARCH-20261020-C4-R1"
      ],
      "provenance_status": "CURRENT_APPROVED_BODY_HASH_MATCH"
    }
  ],
  "research_sections": [
    {
      "heading": "要点と前提",
      "text": "Confidential AI inferenceでは、CPU TEEだけでなくGPU device、driver/firmware、起動状態、workload policyのattestation evidenceを検証し、許可状態と一致した場合だけmodel decryption key、dataset key、API credential等をreleaseする。remote attestationは診断ログではなくsecret release前のpolicy gateとして扱う。\n\nGoogle Cloud Confidential VMのattestationはbootloader、hardware、firmware等のmeasurementを含む。GPU側ではdriver attestationとNVIDIA GPU device attestationが分かれ、NVIDIA Attestation SDKはConfidential Computing対応GPUのlocal/remote attestationを提供する。AI inferenceではmodel executionがGPUへ移るため、VM attestation成功とGPU attestation成功を別evidenceとして保持し、双方を必要とするpolicyならAND条件で判定する。"
    },
    {
      "heading": "運用上の意味・failure",
      "text": "CPU attestationだけでGPU workloadへsecretを渡す、startup時一度のattestationで長寿命credentialを配る、reference value更新を自動許可する、過去tokenをreplay可能にする、といった構成では信頼境界が弱い。hardware measurementが正しくても、hostからguestへの操作面やapplication authorizationが過大なら安全性は成立しない。"
    },
    {
      "heading": "具体的control / implementation implication",
      "text": "release対象を分類し、attestationなしでは取得不能にする。平文secretをimage、startup script、環境変数へ先置きしてからattestationしてはならない。attestation policyはversioned artifactとして管理し、CPU TEE type、expected measurements、GPU model/mode、driver/firmware reference、Secure Boot、workload identity、agent policy等の必須条件を明示する。reference value更新にはchange control、旧新値、適用開始時刻、rollback条件を残す。\n\nnonce/challenge、issuer、audience、expiry等でfreshnessを検証し、過去に成功したtokenのreplayを拒否する。attestation service outage、measurement mismatch、GPU evidence取得失敗はfail-closeを基本とする。break-glassが必要なら通常policyと分離し、承認・期限・監査証跡を持たせる。\n\nNVIDIA Confidential Containersではremote attestation後にKey Broker Service/Attestation Serviceを介してsecretまたはsensitive resourceをreleaseする構造が示されている。attestation resultを記録するだけでなく、resource release pathへ接続することがcontrolの要点である。"
    },
    {
      "heading": "適用境界",
      "text": "Remote attestationは実行環境が期待状態かを検証する仕組みであり、modelの安全性・正確性、prompt injection耐性、application authorizationを証明しない。TEE内部で脆弱なapplicationを正しく起動してもattestationは成功し得る。provider/hardwareごとにclaimsとreference value semanticsも異なるため、Google Cloud Attestation、NVIDIA Remote Attestation Service、AMD SEV-SNP、Intel TDX等を一つのbooleanへ潰さず、raw evidence locator、verifier、policy version、decisionを保存する。"
    },
    {
      "heading": "Counterevidence / limitations",
      "text": "Confidential Computingは攻撃面を消滅させない。firmware、driver、attestation verifier、reference-value supply chain、TEE implementation自体がtrust baseになる。measurement mismatch等のplatform defectもあり得るため、attestation failureを即侵害と断定せず切り分ける必要がある。ただし切り分け中のsecret release許可とは別問題である。provider-operated verifierは独立性要件を満たさない場合があり、第三者/local verifierではreference value配布・revocation・availabilityの運用負担が増える。"
    },
    {
      "heading": "OPEN_UNKNOWN",
      "text": "- CPU TEE、GPU、driver、workload policyの必要十分なclaim組合せ。\n- firmware/driver更新時のreference-value rolloutとfail-close可用性の両立。\n- long-running inferenceの再attestation間隔と既発行secretのrevocation。\n- multi-GPU/NVSwitch構成で全device evidenceを束ねるatomicity。\n- provider verifierと第三者verifierの独立監査境界。"
    },
    {
      "heading": "SAFE_WORDING",
      "text": "「remote attestationに成功すればConfidential AI inferenceは安全」とは断定しない。「CPU/GPU/driver/firmware/workload policy等のattestation evidenceをversioned policyで検証し、許可状態に一致した場合だけsecretや暗号化資産をreleaseすることで、機密AI workloadの実行環境に対する信頼境界を強化する」とする。"
    },
    {
      "heading": "Source / provenance locators",
      "text": "- Google Cloud, Confidential VM attestation: https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/attestation\n- Google Cloud, Confidential VM overview: https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/confidential-vm-overview\n- NVIDIA Attestation SDK: https://docs.nvidia.com/attestation/attestation-client-tools-sdk/latest/gpu_and_switch_attestation.html\n- NVIDIA Confidential Containers Attestation: https://docs.nvidia.com/datacenter/cloud-native/confidential-containers/latest/attestation.html\n- NVIDIA Hopper Remote Attestation Example: https://docs.nvidia.com/attestation/quick-start-guide/latest/attestation-examples/hopper_single_gpu.html"
    }
  ],
  "human_url": "/tools/ea-753-confidential-ai-inference-remote-attestation",
  "markdown_url": "/agent/article.md?path=%2Ftools%2Fea-753-confidential-ai-inference-remote-attestation",
  "json_url": "/agent/article.json?path=%2Ftools%2Fea-753-confidential-ai-inference-remote-attestation"
}