{
  "format": "agent-article-v0",
  "article_id": "EA-065",
  "canonical_url": "/tools/ea-065-ai-records-management-ediscovery",
  "page_type": "reference",
  "claim_ids": [],
  "evidence_ids": [
    "EVD-EA-065-URL-9594B21DCBFA",
    "EVD-EA-065-URL-5B088288FFE9",
    "EVD-EA-065-URL-B8F74260B1EE",
    "EVD-EA-065-URL-0F75EA8F070D",
    "EVD-EA-065-URL-BC920A61C860",
    "EVD-EA-065-URL-2087B04F54A0",
    "EVD-EA-065-URL-D7C707A66DAA",
    "EVD-EA-065-URL-D813B61FD650",
    "EA-065-R11-RESEARCH",
    "EA-065-R11-WEB-READBACK"
  ],
  "unknown_ids": [
    "EA-065-UNKNOWN-01",
    "EA-065-UNKNOWN-CURRENTNESS"
  ],
  "operational_object_ids": [],
  "provider": [
    "Microsoft",
    "OpenAI"
  ],
  "models": [],
  "plans": [],
  "surfaces": [],
  "regions": [],
  "current_status": "UNKNOWN",
  "observation_status": "SPEC_ONLY",
  "freshness": "UNKNOWN",
  "last_verified": null,
  "supersedes": [],
  "superseded_by": [],
  "provenance": [
    "COMM-061-EA0065",
    "SRC-EA-065-URL-9594B21DCBFA",
    "SRC-EA-065-URL-5B088288FFE9",
    "SRC-EA-065-URL-B8F74260B1EE",
    "SRC-EA-065-URL-0F75EA8F070D",
    "SRC-EA-065-URL-BC920A61C860",
    "SRC-EA-065-URL-2087B04F54A0",
    "SRC-EA-065-URL-D7C707A66DAA",
    "SRC-EA-065-URL-D813B61FD650",
    "WEBBODY-EA0065-R8-EEBE0168"
  ],
  "article_sources": [
    {
      "id": "EVD-EA-065-URL-9594B21DCBFA",
      "evidence_class": "RESEARCH",
      "title": "- Author/Org: U.S. National Archives and Records Administration (NARA)",
      "href": "https://www.archives.gov/records-mgmt/memos/ac-11-2026",
      "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-065-URL-9594B21DCBFA"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "- Canonical URL: https://www.archives.gov/records-mgmt/memos/ac-11-2026"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-065-URL-5B088288FFE9",
      "evidence_class": "RESEARCH",
      "title": "- Author/Org: NARA",
      "href": "https://www.archives.gov/files/records-mgmt/policy/nara-fra-ai-guidance.pdf",
      "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-065-URL-5B088288FFE9"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "- Canonical URL: https://www.archives.gov/files/records-mgmt/policy/nara-fra-ai-guidance.pdf"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-065-URL-B8F74260B1EE",
      "evidence_class": "RESEARCH",
      "title": "- Author/Org: Microsoft",
      "href": "https://learn.microsoft.com/en-us/purview/ai-microsoft-purview",
      "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-065-URL-B8F74260B1EE"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "- Canonical URL: https://learn.microsoft.com/en-us/purview/ai-microsoft-purview"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-065-URL-0F75EA8F070D",
      "evidence_class": "RESEARCH",
      "title": "- Author/Org: Microsoft",
      "href": "https://learn.microsoft.com/purview/edisc-search-copilot-data",
      "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-065-URL-0F75EA8F070D"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "- Canonical URL: https://learn.microsoft.com/purview/edisc-search-copilot-data"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-065-URL-BC920A61C860",
      "evidence_class": "RESEARCH",
      "title": "- Author/Org: Microsoft",
      "href": "https://learn.microsoft.com/en-us/purview/retention-policies-copilot",
      "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-065-URL-BC920A61C860"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "- Canonical URL: https://learn.microsoft.com/en-us/purview/retention-policies-copilot"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-065-URL-2087B04F54A0",
      "evidence_class": "RESEARCH",
      "title": "- Author/Org: Microsoft",
      "href": "https://learn.microsoft.com/en-us/purview/edisc-search-export",
      "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-065-URL-2087B04F54A0"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "- Canonical URL: https://learn.microsoft.com/en-us/purview/edisc-search-export"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-065-URL-D7C707A66DAA",
      "evidence_class": "RESEARCH",
      "title": "- Author/Org: OpenAI",
      "href": "https://help.openai.com/en/articles/9261474",
      "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-065-URL-D7C707A66DAA"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "- Canonical URL: https://help.openai.com/en/articles/9261474"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-065-URL-D813B61FD650",
      "evidence_class": "RESEARCH",
      "title": "- Author/Org: Administrative Office of the U.S. Courts / Rules process",
      "href": "https://www.uscourts.gov/rules-policies/current-rules-practice-procedure",
      "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-065-URL-D813B61FD650"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "- Canonical URL: https://www.uscourts.gov/rules-policies/current-rules-practice-procedure"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EA-065-R11-RESEARCH",
      "evidence_class": "RESEARCH",
      "title": "EA-065 監査済みresearch / canonical revision binding",
      "supports": [
        "### F1 — “AI-generated” is not a record classification\nNARAの2026 guidanceでは、AI inputs/outputs/audit trails/software等はrecord評価の対象になり得るが、AI利用そのものではrecord化しない。official businessでの利用、decision-makingへの依拠、circulation、systemへのcapture/control、business/legal obligation等が判定材料になる。したがって将来版で「AI会話は全部record」「AI logは全部保存」と一般化してはならない。"
      ],
      "limit": "canonical監査・revision bindingの存在を示す。外部ソースcurrentnessが本文またはcontrol-planeで確定していない場合は、currentnessをUNKNOWNのまま保持する。",
      "origin_class": "AUDITED_CANONICAL_RESEARCH_PACKET",
      "source_or_experiment_ids": [
        "COMM-061-EA0065"
      ],
      "supporting_locators": [
        "COMM-061-EA0065",
        "EA-065-R5-LEGACY-MIGRATION-20261008-R1",
        "content/staging/EA-065.md"
      ],
      "scope": "監査済みEA本文・canonical revision・既存Source Graph relationの範囲。欠損currentnessは推測で閉じない。",
      "provenance_status": "AUDIT_PASS_CANONICAL_BINDING"
    },
    {
      "id": "EA-065-R11-WEB-READBACK",
      "evidence_class": "IMPL",
      "title": "EA-065 Web body identity / canonical readback",
      "supports": [
        "canonical staging bodyのSHA-256一致とR9 canonical ledger readbackをProduction projectionの入力として検証する。"
      ],
      "limit": "body identity/readbackは本文claim自体の真偽や外部ソースの最新性を追加保証しない。",
      "checked_at": null,
      "origin_class": "IMPLEMENTATION_READBACK",
      "source_or_experiment_ids": [
        "WEBBODY-EA0065-R8-EEBE0168"
      ],
      "supporting_locators": [
        "WEBBODY-EA0065-R8-EEBE0168",
        "content/staging/EA-065.md",
        "EA-065-R5-LEGACY-MIGRATION-20261008-R1"
      ],
      "scope": "R9/R10 canonical readback後のProduction生成時body identity。",
      "provenance_status": "CANONICAL_READBACK_BODY_HASH_MATCH"
    }
  ],
  "research_sections": [
    {
      "heading": "WHY_THIS_MATTERS_TO_UPDATE",
      "text": "Enterprise AIでは、prompt/response、AI生成成果物、Agent実行ログ、tool result、memory、監査ログなどが複数の保存面に分散する。これらを一律に「AIログ」として保存・削除すると、(a) business recordか否か、(b) retention/records schedule、(c) legal hold、(d) eDiscoveryの検索・export、(e) user-visible deletion、(f) provider-side compliance log retentionを混同する。本項は、AI利用物を記録管理とdiscoveryへ接続する際に、何をrecord/evidenceとして扱い、どのrepositoryとlifecycleを追跡し、hold・search・export・deletionをどう分離するかの更新候補を確立する。"
    },
    {
      "heading": "SCOPE_ROUTING",
      "text": "- Enterprise AI Core: artifact inventory / provenance / retention / deletion / audit\n- Governance: records classification / retention schedule / legal hold / investigation\n- Interoperability: provider compliance export → eDiscovery / archive / SIEM / DLP\n- Provider mapping: Microsoft Purview / OpenAI Compliance Platform（current product claim）\n- Organization overlay: records policy、legal/investigation trigger、custodian、jurisdiction、case procedure\n- Non-LLM component: mailbox/compliance store、archive、eDiscovery index/export、case management\n- 法域注意: NARA/FRAおよび米国民事訴訟の資料は米国文脈の一次根拠であり、他法域への直接外挿はしない。"
    },
    {
      "heading": "CLAIM_OR_RESEARCH_QUESTION",
      "text": "1. AI会話・AI生成物・audit trail・Agent実行は、AI由来であることだけでrecordになるのか。\n2. records managementとeDiscovery/legal holdを同一概念として扱えるか。\n3. AI artifactの「保持」「検索可能」「export可能」「削除済み」は同じ状態か。\n4. current enterprise productsはどのAI interaction surfaceをcompliance/eDiscoveryへ接続でき、どこにavailability/lifecycle gapがあるか。\n5. 将来版Enterprise AI Coreは何をartifact単位で記録すべきか。"
    },
    {
      "heading": "EVIDENCE TABLE",
      "text": "### E1 — primary government guidance / support+limit\n- Exact title: Guidance on Applying the Federal Records Act to Artificial Intelligence Materials / AC 11.2026\n- Author/Org: U.S. National Archives and Records Administration (NARA)\n- Canonical URL: https://www.archives.gov/records-mgmt/memos/ac-11-2026\n- Publication / update / access: 2026-08-21 / last reviewed 2026-09-03 / accessed 2026-09-15\n- Version: AC 11.2026\n- Supporting locator: page summary, Part I/II description\n- Evidence label: VERIFIED_SPEC / SOURCE_REPORT\n- Relation: support + limit\n- Establishes: FRAのrecord definitionをAI inputs, outputs, data, audit trails, software等へ適用し、AI-related federal recordsの廃棄はNARA-approved scheduleに従う。しかも同guidanceはAI governance、e-discovery、privacy、security、ethical useのpolicyを定めないと明示する。\n\n\n### E2 — primary government guidance / support+counter\n- Exact title: Guidance on Applying the Federal Records Act to Artificial Intelligence Materials (memorandum PDF)\n- Author/Org: NARA\n- Canonical URL: https://www.archives.gov/files/records-mgmt/policy/nara-fra-ai-guidance.pdf\n- Publication / access: 2026-08-21 / 2026-09-15\n- Version: AC 11.2026 memorandum\n- Supporting locator: pp.1–5, especially federal record definition; pp.2–4 examples; Part II\n- Evidence label: VERIFIED_SPEC / SOURCE_REPORT\n- Relation: support + counter + limit\n- Establishes: AI platformを使った事実だけではfederal recordは自動生成されない。official business、decision-making、circulation、agency systemへのcapture/use等の状況が重要。mere retention of an audit trail does not itself establish record status。AI outputのscheduleはAIだから特別になるのではなく、例えばemailならemail、final reportならreportに適用されるscheduleが基準となる。\n\n\n### E3 — current product documentation / support\n- Exact title: Microsoft Purview data security and compliance protections for generative AI apps\n- Author/Org: Microsoft\n- Canonical URL: https://learn.microsoft.com/en-us/purview/ai-microsoft-purview\n- Update / access: current page / accessed 2026-09-15\n- Product/version: Microsoft Purview current service documentation; PRODUCT_VOLATILE\n- Supporting locator: sections “Microsoft Purview supports compliance management for AI apps” and “Auditing and AI interactions”\n- Evidence label: VERIFIED_SPEC / PRODUCT_VOLATILE\n- Relation: support + availability context\n- Establishes: supported AI interactions can be subject to auditing, communication compliance, eDiscovery content search, retention/deletion; prompts/responses can be captured in unified audit log with service/file references. Supported app categories are explicitly enumerated and therefore coverage is app/mode dependent。\n\n\n### E4 — current product documentation / support+counter\n- Exact title: Search for and delete AI application data in eDiscovery\n- Author/Org: Microsoft\n- Canonical URL: https://learn.microsoft.com/purview/edisc-search-copilot-data\n- Update / access: current page / accessed 2026-09-15\n- Product/version: Microsoft Purview eDiscovery current documentation; PRODUCT_VOLATILE\n- Supporting locator: search tips; Copilot memory notes; hold/retention removal before deletion\n- Evidence label: VERIFIED_SPEC / PRODUCT_VOLATILE\n- Relation: support + counter + limit\n- Establishes: AI dataをeDiscoveryで検索・削除する経路がある一方、Copilot inferred memoryは別item classであり、conversation/messageの削除はassociated memoryを削除しない。holdまたはretention policyが対象mailboxに残ると削除対象dataはretainされる。\n\n\n### E5 — current product documentation / support+limit\n- Exact title: Learn about retention for Copilot and AI apps\n- Author/Org: Microsoft\n- Canonical URL: https://learn.microsoft.com/en-us/purview/retention-policies-copilot\n- Update / access: page current at access 2026-09-15\n- Product/version: Microsoft Purview retention current documentation; PRODUCT_VOLATILE\n- Supporting locator: “How retention works with AI apps”\n- Evidence label: VERIFIED_SPEC / PRODUCT_VOLATILE\n- Relation: support + limit\n- Establishes: compliance用AI message dataはbackend mailbox hidden folderへ保存され、retention jobとSubstrateHoldsを経由する。Litigation Hold/eDiscovery hold等がある場合permanent deletionはsuspendされる。user-visible app stateとcompliance retention stateは一致しないことがあり、permanent deletionまでの間はeDiscovery search可能。\n\n\n### E6 — current product documentation / support+counter\n- Exact title: Export search results in eDiscovery\n- Author/Org: Microsoft\n- Canonical URL: https://learn.microsoft.com/en-us/purview/edisc-search-export\n- Update / access: current page / accessed 2026-09-15\n- Product/version: Microsoft Purview eDiscovery current documentation; PRODUCT_VOLATILE\n- Supporting locator: Export search results, indexed/partially indexed options\n- Evidence label: VERIFIED_SPEC / PRODUCT_VOLATILE\n- Relation: support + counter\n- Establishes: search/exportではindexed itemsとpartially/unindexed itemsの扱いを分離でき、search-timeとexport-timeの設定は独立する。したがって「検索で見える／統計に出る」と「同じ条件で完全にexportされる」は同義ではない。\n\n\n### E7 — current provider documentation / support+limit\n- Exact title: OpenAI Compliance Platform for Enterprise and Edu customers\n- Author/Org: OpenAI\n- Canonical URL: https://help.openai.com/en/articles/9261474\n- Update / access: current page / accessed 2026-09-15\n- Product/version: Compliance Logs Platform; new conversation logs system released 2026-03-05; old stateful route removed 2026-06-05; PRODUCT_VOLATILE\n- Supporting locator: Data retention with the Compliance API; Deprecation notice; Partner integrations\n- Evidence label: VERIFIED_SPEC / PRODUCT_VOLATILE\n- Relation: support + limit + interoperability\n- Establishes: Compliance Logs Platformは30日retainし、それ以上のretentionにはconsumer側のcontinuous download/retentionが必要。eDiscovery/DLP/security provider integrationが用意される。Compliance Platformへの接続が、そのまま長期archiveまたはlegal hold機能であるとは言えない。\n\n\n### E8 — primary U.S. eDiscovery context / context+limit\n- Exact title: Federal Rules of Civil Procedure — Rule 34/37 electronic discovery framework (official U.S. Courts materials)\n- Author/Org: Administrative Office of the U.S. Courts / Rules process\n- Canonical URL: https://www.uscourts.gov/rules-policies/current-rules-practice-procedure\n- Publication/update/access: current rules framework / accessed 2026-09-15\n- Version: current federal civil procedure framework; jurisdiction-specific\n- Supporting locator: Rule 34 electronically stored information; Rule 37(e) preservation/loss framework\n- Evidence label: VERIFIED_SPEC / SOURCE_REPORT\n- Relation: context + limit\n- Establishes: U.S. federal civil discovery treats ESI broadly and preservation/loss obligations are litigation-context questions. AI-specific record statusを自動決定する規則としては扱わない。"
    },
    {
      "heading": "FINDINGS",
      "text": "### F1 — “AI-generated” is not a record classification\nNARAの2026 guidanceでは、AI inputs/outputs/audit trails/software等はrecord評価の対象になり得るが、AI利用そのものではrecord化しない。official businessでの利用、decision-makingへの依拠、circulation、systemへのcapture/control、business/legal obligation等が判定材料になる。したがって将来版で「AI会話は全部record」「AI logは全部保存」と一般化してはならない。\n\n\n### F2 — Records Management と eDiscovery は別レイヤ\nNARA自身がAC 11.2026の射程からe-discoveryを除外している。records managementは何を公式記録としてどのscheduleで保持・廃棄するかの問題であり、eDiscovery/legal holdはlitigation/investigation時にresponsive ESIをpreserve/search/review/exportする問題である。両者はrepository、retention、provenanceを共有し得るが同義ではない。\n\n\n### F3 — AI artifactはrepository/lifecycle単位で管理する必要がある\ncurrent product evidenceでは、prompt/response、memory、audit/compliance logが別surfaceに存在する。Microsoftではconversation削除とmemory削除が分離し、compliance copyはhidden mailbox/hold経路を持つ。OpenAI Compliance Logs Platformは30日retentionで、長期保持はdownstream consumerが担う。よって「チャットを消した」「providerにlogがある」だけではrecords/eDiscovery要件を閉じられない。\n\n\n### F4 — hold / retention / deletion / search / export は別状態\nMicrosoftの仕様ではholdやretention policyがpermanent deletionを止め得る。またsearch/export時のindexed/partially indexed itemの扱いは独立する。従って最低でも次を別fieldで持つべきである。\n- record_status\n- retention_schedule / retention_until\n- hold_status / hold_scope\n- production_visibility\n- compliance_searchability\n- exportability / export_scope\n- logical_delete_status\n- permanent_delete_status / expected purge window\n\n\n### F5 — legal holdはprovider UIではなくorganization processとして設計する\n特定providerがeDiscovery/retention APIを提供しても、trigger判定、custodian/repository scope、hold authorization、release authorization、chain of custodyはorganization overlayである。provider機能の存在を「法的保全が成立した」証明へ外挿しない。\n\n\n### F6 — AI provenanceはeDiscovery価値を上げるが、存在しないmetadataを推測しない\nAI interactionにactor、timestamp、app/agent、model/config revision、source/tool link、artifact IDが残れば再現・説明・reviewが容易になる。しかしproviderがsystem prompt、internal reasoning、非公開telemetry、完全なmodel revision等を提供するとは限らない。public docsで明示されないsurfaceはUNKNOWNのまま扱う。"
    },
    {
      "heading": "LIMITS / COUNTEREVIDENCE / FAILURE ENVELOPE",
      "text": "- NARA guidanceはU.S. federal recordsへの一次根拠であり、民間企業・日本その他法域のrecords lawへ直接外挿しない。\n- NARAはe-discovery policyを定めていないため、records scheduleだけでlegal hold/discoveryを完結させない。\n- mere retention / audit trail existence ≠ record。逆にrecord policy未整備でもbusiness/legal needによりrecordとなる場合がある。\n- Provider public docs ≠ specific tenant configuration。license、region、app category、connector、permission、admin settingによりavailabilityは変わる。\n- user-visible deletion ≠ compliance-store deletion ≠ memory deletion ≠ downstream archive deletion。\n- eDiscovery search hit ≠ complete export。index状態、export setting、DLP/encryption、source coverage等の影響を受ける。\n- compliance API/log retention ≠ legal hold。OpenAI current docsでは30日を超える保持にはconsumer側収集が必要。\n- provider system内部のCoT/internal telemetry等をdiscoverableと推定しない。\n- “all AI activity”という表現は、明示されたcoverage matrixなしでは使用しない。"
    },
    {
      "heading": "AVAILABILITY_GATES",
      "text": "### Microsoft Purview\n- supported AI app/category gate\n- Purview/eDiscovery licensing and role/permission gate\n- mailbox/location/source coverage\n- retention/hold configuration\n- cloud/region/government environment differences\n- index/export settings and item indexability"
    },
    {
      "heading": "OpenAI Compliance Platform",
      "text": "- ChatGPT Enterprise / Edu workspace gate\n- workspace/admin permission/API-key scope gate\n- Compliance Logs Platform schema/version gate\n- 30-day provider retention vs downstream retention architecture\n- partner integration coverage"
    },
    {
      "heading": "IMPLEMENTATION_STATE",
      "text": "- Records/retention/eDiscovery control自体は成熟した非LLM enterprise controlとして既に存在する。\n- AI interaction integrationもMicrosoft Purview/OpenAI Compliance Platform等で実装済み。\n- ただしcross-providerで共通の「AI record schema」「universal hold API」「全artifact自動capture」は確認できない。\n- current practical implementationは provider capture/export + organization archive/eDiscovery/records system の連携型と見るのが安全。"
    },
    {
      "heading": "BENCHMARK_OR_EVAL",
      "text": "性能benchmarkは本項の中心ではない。必要なのはcoverage/retrievability/hold/deletion verification。\n\n\n### TEST_PROFILE — AI_RECORDS_EDISCOVERY_CANARY_V1（未実施）\n認証済みtenant操作を要するため本RUNでは実施しない。\n1. synthetic canaryのみを用い、prompt、response、generated file、tool result、agent execution event、memory、audit/compliance logを別IDで生成。\n2. 各artifactについてauthoritative repositoryとcapture timestampを記録。\n3. eDiscovery/searchでartifactごとのhit/miss、index delay、metadataを観測。\n4. authorized test tenantでretention/holdを付与し、user/app側削除とcompliance側残存を比較。\n5. conversation削除後にmemory、audit、downstream exportが残るか個別確認。\n6. exportし、artifact ID、timestamp、custodian/source、file/hash等のpreservation metadataを確認。\n7. hold release後、documented purge windowを経て再検索。\n8. false conclusion禁止: app非表示、404、検索hitなし、agent非想起の単独結果をphysical purge/comprehensive deletionの証明にしない。"
    },
    {
      "heading": "SECURITY / PRIVACY / AUDIT / COST / OPERABILITY",
      "text": "### Security\n- eDiscovery/compliance APIは高権限面。least privilege、role separation、case-scoped access、admin/API key protectionが必要。\n- exportは新しいsensitive copyを作るため、export destinationのencryption/access/retentionも管理対象。\n\n\n### Privacy\n- “preserve everything”はprivacy/minimizationと衝突し得る。record/hold basisとscopeを明示し、法的・業務的に必要なartifactを対象化する。\n- sensitive prompt/responseをreview/exportする場合、reviewer accessとsecondary useを管理する。\n\n\n### Audit\n- hold create/change/release、search query、export、delete、role assignmentを監査証跡化する。\n- artifact capture provenanceとcase action provenanceを分離して保持する。\n\n\n### Cost\n- providerからのcontinuous log export、long-term archive、indexing、review/export storageは追加costとなる。OpenAIの30日retentionを超えて保存する設計では特にdownstream storage/ingestion costを見積もる。\n\n\n### Operability\n- providerごとの保存面を一枚のmapに寄せることが運用上の要点。product UI名称よりartifact/repository/retention/hold/search/export/deletion stateを正本にする。"
    },
    {
      "heading": "SAFE_WORDING_FOR_FUTURE_BOOK",
      "text": "「AI会話・生成物・Agent実行は、AIで生成されたという理由だけで一律に記録または廃棄対象になるわけではない。記録性は組織の業務・法令・保存方針に従ってartifact単位で判断し、legal holdやeDiscoveryでは、会話、memory、監査ログ、成果物などの実際の保存先を把握して保全・検索・export範囲を確認する。ユーザー画面からの削除、検索不能、provider側retention終了は同じ事象ではない。製品のcompliance/eDiscovery連携は有力な実装手段だが、特定tenantでの適用範囲、hold状態、index/export coverageを別途検証する必要がある。」"
    },
    {
      "heading": "UPDATE_WORK_HANDOFF",
      "text": "将来update Workでは、本項を単独の法律解説にせず、EA-064 Memory Lifecycle、EA-058 configuration management、EA-055 Agent Replay/Forensics、EA-056 AI Asset Inventoryと接続する。\n\n\n### Core candidate: AI RECORD EVIDENCE MAP（PROPOSAL / INFERENCE）\n最低field:\n- artifact_id / artifact_type\n- business_record_status + basis + authority\n- authoritative_repository / provider / tenant\n- custodian / actor / agent / service principal\n- created_at / captured_at\n- source/provenance / parent-child relation\n- model/config/tool/source revision where observable\n- retention_schedule / retention_until\n- hold_status / hold_id / hold_scope\n- search/index status\n- export format / export availability / export timestamp / checksum if available\n- user-visible deletion status\n- compliance-store deletion status / expected purge window\n- downstream copies / archive destinations\n- evidence source / source version / as-of date\n\n\n### Core candidate: hold/readiness flow（PROPOSAL）\nInventory repositories → classify record/retention basis → legal/investigation triggerでcustodian/repository scope確定 → destructive lifecycleをsuspend/hold → search/index coverage確認 → export/preserve metadata → review/production → authorized release → documented retention/deletionへ復帰。\n\n\n### Provider Mappingで再確認するもの\n- Microsoft: supported AI app matrix、Copilot memory、hidden mailbox/retention、hold/search/export semantics、license/role/cloud gates。\n- OpenAI: Compliance Logs event/schema、current retention、workspace plan/admin gate、deletion semantics、eDiscovery partner integration。\n- 他provider: equivalent compliance export / audit / retention / legal hold pathの有無は別採集。存在を推測しない。"
    },
    {
      "heading": "OPEN_UNKNOWN",
      "text": "1. provider横断でAI conversation/memory/agent traceを共通表現するrecords/eDiscovery標準 — UNKNOWN。\n2. 各providerの全AI surfaceをlegal hold対象へ直接固定できるか — provider別UNKNOWN。\n3. system prompt、hidden orchestration state、internal reasoning/telemetryのdiscoverability — explicit documentationがない限りUNKNOWN。\n4. memory/vector index/derived summaryが元conversationのholdを自動継承するか — provider別UNKNOWN。\n5. 日本その他法域でのAI-specific records/eDiscovery要件 — jurisdiction-specific採集が必要。\n6. downstream connector/archiveへexport後のdeletion propagation — integration別UNKNOWN。"
    },
    {
      "heading": "STOP_CONDITION",
      "text": "以下が一次/current資料で揃ったためEA-065をbounded closeする。\n- AI materialのrecord statusが「AI由来」だけで自動決定されない一次根拠。\n- records managementとeDiscoveryを分離すべき一次根拠。\n- current enterprise AI interactionについてhold/retention/search/export/deleteが別stateである実装証拠。\n- MicrosoftとOpenAIのcurrent provider mapping例、availability gate、counterevidence。\n- 将来Coreがそのまま採否判断できるAI Record Evidence Map、TEST_PROFILE、OPEN_UNKNOWN。\n\n\nBOUNDED_CLOSE: EA-065\n\n---"
    },
    {
      "heading": "AUDITED_BOUNDARY",
      "text": "AI origin does not itself determine record status. Records management, retention schedule, legal hold, compliance searchability, exportability, user-visible deletion and permanent deletion are separate. NARA AC 11.2026 excludes e-discovery policy. OpenAI compliance logs retention is not legal hold; current access patterns must remain distinct. GX relation is RELATED_GX_NOTE_ONLY."
    }
  ],
  "human_url": "/tools/ea-065-ai-records-management-ediscovery",
  "markdown_url": "/agent/article.md?path=%2Ftools%2Fea-065-ai-records-management-ediscovery",
  "json_url": "/agent/article.json?path=%2Ftools%2Fea-065-ai-records-management-ediscovery"
}