{ "format": "agent-article-v0", "article_id": "EA-064", "canonical_url": "/tools/ea-064-ai-data-memory-lifecycle", "page_type": "reference", "claim_ids": [], "evidence_ids": [ "EVD-EA-064-URL-2ACA780AF5AA", "EVD-EA-064-URL-464C77F21263", "EVD-EA-064-URL-E51F0E399A8F", "EVD-EA-064-URL-E1D8694DCCFC", "EVD-EA-064-URL-C9D674FF5EF1", "EVD-EA-064-URL-CFB8F2CA4245", "EVD-EA-064-URL-7C4686E15F54", "EVD-EA-064-URL-48E2615E119F", "EA-064-R11-RESEARCH", "EA-064-R11-WEB-READBACK" ], "unknown_ids": [ "EA-064-UNKNOWN-01", "EA-064-UNKNOWN-CURRENTNESS" ], "operational_object_ids": [], "provider": [ "OpenAI", "Google" ], "models": [], "plans": [], "surfaces": [], "regions": [], "current_status": "UNKNOWN", "observation_status": "SPEC_ONLY", "freshness": "UNKNOWN", "last_verified": null, "supersedes": [], "superseded_by": [], "provenance": [ "COMM-060-EA0064", "SRC-EA-064-URL-2ACA780AF5AA", "SRC-EA-064-URL-464C77F21263", "SRC-EA-064-URL-E51F0E399A8F", "SRC-EA-064-URL-E1D8694DCCFC", "SRC-EA-064-URL-C9D674FF5EF1", "SRC-EA-064-URL-CFB8F2CA4245", "SRC-EA-064-URL-7C4686E15F54", "SRC-EA-064-URL-48E2615E119F", "WEBBODY-EA0064-R8-B6E24E34" ], "article_sources": [ { "id": "EVD-EA-064-URL-2ACA780AF5AA", "evidence_class": "RESEARCH", "title": "author/org: OpenAI", "href": "https://developers.openai.com/api/docs/guides/your-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-064-URL-2ACA780AF5AA" ], "supporting_locators": [ "EVIDENCE TABLE", "canonical URL: https://developers.openai.com/api/docs/guides/your-data" ], "scope": "EVIDENCE TABLE", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-064-URL-464C77F21263", "evidence_class": "RESEARCH", "title": "author/org: Google Cloud", "href": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/scale/memory-bank/revisions", "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-064-URL-464C77F21263" ], "supporting_locators": [ "EVIDENCE TABLE", "canonical URL: https://docs.cloud.google.com/gemini-enterprise-agent-platform/scale/memory-bank/revisions" ], "scope": "EVIDENCE TABLE", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-064-URL-E51F0E399A8F", "evidence_class": "RESEARCH", "title": "author/org: Google Cloud", "href": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/scale/memory-bank/setup", "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-064-URL-E51F0E399A8F" ], "supporting_locators": [ "EVIDENCE TABLE", "canonical URL: https://docs.cloud.google.com/gemini-enterprise-agent-platform/scale/memory-bank/setup" ], "scope": "EVIDENCE TABLE", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-064-URL-E1D8694DCCFC", "evidence_class": "RESEARCH", "title": "author/org: Amazon Web Services", "href": "https://docs.aws.amazon.com/bedrock/latest/userguide/agents-configure-memory.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-064-URL-E1D8694DCCFC" ], "supporting_locators": [ "EVIDENCE TABLE", "canonical URL: https://docs.aws.amazon.com/bedrock/latest/userguide/agents-configure-memory.html" ], "scope": "EVIDENCE TABLE", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-064-URL-C9D674FF5EF1", "evidence_class": "RESEARCH", "title": "author/org: Amazon Web Services", "href": "https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/memory-create-a-memory-store.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-064-URL-C9D674FF5EF1" ], "supporting_locators": [ "EVIDENCE TABLE", "canonical URL: https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/memory-create-a-memory-store.html" ], "scope": "EVIDENCE TABLE", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-064-URL-CFB8F2CA4245", "evidence_class": "RESEARCH", "title": "author/org: NIST", "href": "https://nvlpubs.nist.gov/nistpubs/CSWP/NIST.CSWP.01162020.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-064-URL-CFB8F2CA4245" ], "supporting_locators": [ "EVIDENCE TABLE", "canonical URL: https://nvlpubs.nist.gov/nistpubs/CSWP/NIST.CSWP.01162020.pdf" ], "scope": "EVIDENCE TABLE", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-064-URL-7C4686E15F54", "evidence_class": "RESEARCH", "title": "author/org: Ramaswamy Chandramouli (NIST), Eric A. Hibbard (Samsung Semiconductor) / NIST", "href": "https://csrc.nist.gov/pubs/sp/800/88/r2/final", "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-064-URL-7C4686E15F54" ], "supporting_locators": [ "EVIDENCE TABLE", "canonical URL: https://csrc.nist.gov/pubs/sp/800/88/r2/final" ], "scope": "EVIDENCE TABLE", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-064-URL-48E2615E119F", "evidence_class": "RESEARCH", "title": "author/org: Md Nayem Uddin, Kumar Shubham, Eduardo Blanco, Chitta Baral, Gengyu Wang / Findings of ACL 2026", "href": "https://aclanthology.org/2026.findings-acl.1337/", "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-064-URL-48E2615E119F" ], "supporting_locators": [ "EVIDENCE TABLE", "canonical URL: https://aclanthology.org/2026.findings-acl.1337/" ], "scope": "EVIDENCE TABLE", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EA-064-R11-RESEARCH", "evidence_class": "RESEARCH", "title": "EA-064 監査済みresearch / canonical revision binding", "supports": [ "- VERIFIED_SPEC: “memory”はprovider内でも単一artifactではない。OpenAIはabuse log/application state、Googleはcurrent Memory/revision、AWSはraw event/session summary等を別レイヤとして扱う。よってEnterprise AI Coreはartifact classごとにlifecycleを追う必要がある。\n- VERIFIED_SPEC: deleteは即時・全面消去と同義ではない。Google Memory Bankは親memory削除後もrevisionを最大48時間参照・rollback可能。OpenAIでも一部stateful objectは削除後server deletionまで30日とされる。\n- VERIFIED_SPEC: TTLの作用範囲と更新規則は製品固有。Googleではupdate/generateがexpirationを再計算し得る。AWS AgentCoreではretention設定変更が既存eventへ遡及しない。したがって「retentionを30日に変更したから既存dataも30日以内に消える」とは一般化できない。\n- INFERENCE: summarization/extraction/consolidationは派生data生成である。raw sessionを削除してもderived summary/fact/revision/indexが連鎖削除されるかは個別仕様または実測で確認すべきである。\n- PROPOSAL: Coreに `AI Memory Lifecycle Matrix / Artifact Deletion Graph` を置き、少なくとも artifact_type / source / derived_from / scope(user, session, agent, org) / store / create_time / update_time / TTL / retention_rule / delete_method / delete_requested_at / logical_inaccessibility_at / expected_purge_at / residual_window / external_copy / evidence を管理する。\n- PROPOSAL: “reset”は曖昧語として禁止せず、必ずdestructive scope manifestを伴わせる。reset対象にsession、durable memory、revision、vector/index、files、logs、tool/MCP downstream stateのどこまでを含むかを明示する。" ], "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-060-EA0064" ], "supporting_locators": [ "COMM-060-EA0064", "EA-064-R5-LEGACY-MIGRATION-20261008-R1", "content/staging/EA-064.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-064-R11-WEB-READBACK", "evidence_class": "IMPL", "title": "EA-064 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-EA0064-R8-B6E24E34" ], "supporting_locators": [ "WEBBODY-EA0064-R8-B6E24E34", "content/staging/EA-064.md", "EA-064-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の「memory」は単一ストアではない。raw conversation/session event、要約、抽出fact、長期memory、revision、embedding/vector index、application state、abuse/audit log、backup/replica、外部tool/MCP側stateなどが別ライフサイクルを持ち得る。したがって「memoryを消した」「resetした」を一語で扱うと、active read pathから消えただけなのか、派生物・revision・ログ・復旧用コピーまで消えたのかを誤認する。本項は、remember/forgetの機能説明ではなく、生成から残存検証までをartifact別に管理する運用層を追加するための証拠棚である。" }, { "heading": "SCOPE_ROUTING", "text": "Enterprise AI Core / provider mapping / organization overlay / governance / privacy / security / audit / implementation / benchmark。model-weight unlearningそのものは本Packetの主対象外とし、application/session/memory artifactの削除からweight-level unlearningを推論しない。" }, { "heading": "CLAIM_OR_RESEARCH_QUESTION", "text": "1. Enterprise AIのdata/memory lifecycleは、create/ingest → derive/extract/summarize → persist → retrieve/use → update/consolidate → expire/TTL → logical delete/reset → recovery/replica purge → residual verificationの複数段階として扱うべきか。\n2. delete/resetはどのartifactを対象とし、どの時点で何が不可視・削除・回復不能になるのか。\n3. TTL、論理削除、behavioral forgetting、storage sanitizationは同義ではないことを、現行provider仕様と研究でどこまで確立できるか。\n4. 後続Workが安全に「忘却」を検証するためのprovider-neutral TEST_PROFILEはどう設計すべきか。" }, { "heading": "EVIDENCE TABLE", "text": "1) source class: official product documentation\nexact title: Data controls in the OpenAI platform\nauthor/org: OpenAI\ncanonical URL: https://developers.openai.com/api/docs/guides/your-data\npublication/update/access date: update date not stated / accessed 2026-09-14\nversion/release/API revision/model ID: current API data-controls documentation as-of 2026-09-14\nsupporting locator: “Types of data stored with the OpenAI API”; “Data retention controls for abuse monitoring”; “Storage requirements and retention controls per endpoint”; endpoint-specific notes for Responses / Assistants / Threads / Vector Stores / Files\nevidence label: VERIFIED_SPEC + PRODUCT_VOLATILE\nrelation: support + limit + availability gate\nfinding: abuse monitoring logsとapplication stateは別カテゴリ。標準abuse monitoringは最大30日。ZDR/MAMは承認制で、ZDRはResponses/Chat Completionsのstoreをfalse扱いにする一方、ZDR非適格endpoint/capabilityはapplication stateを保持し得る。Conversations等はuntil deleted、Assistants/Threads/Vector Stores等はuntil deletedかつ削除後server deletionまで30日と記載。Filesはmanual deleteまたはexpires_afterを持つ。画像/fileの一部safety例外やRemote MCP第三者ポリシーも存在。\n\n\n2) source class: official product documentation\nexact title: Memory revisions | Gemini Enterprise Agent Platform | Google Cloud Documentation\nauthor/org: Google Cloud\ncanonical URL: https://docs.cloud.google.com/gemini-enterprise-agent-platform/scale/memory-bank/revisions\npublication/update/access date: accessed 2026-09-14\nversion/release/API revision/model ID: current Agent Platform Memory Bank documentation as-of 2026-09-14\nsupporting locator: “Lifecycle of a memory and its revisions”; “Memory deletion”; “Revision expiration”\nevidence label: VERIFIED_SPEC + PRODUCT_VOLATILE\nrelation: support + counter/limit\nfinding: Memory BankはMemoryとimmutable MemoryRevisionを分離し、revisionは既定で有効。Memory削除時、現行Memory resourceは消えるがdelete mutationのrevisionが作られ、child revisionsは親削除後最大48時間参照できrollbackも可能。通常のrevision TTLは既定365日。active object deletionとhistorical state消去は同一時点ではない。\n\n\n3) source class: official product documentation\nexact title: Set up Memory Bank | Gemini Enterprise Agent Platform | Google Cloud Documentation\nauthor/org: Google Cloud\ncanonical URL: https://docs.cloud.google.com/gemini-enterprise-agent-platform/scale/memory-bank/setup\npublication/update/access date: accessed 2026-09-14\nversion/release/API revision/model ID: Agent Platform SDK example google-cloud-aiplatform>=1.111.0; current docs as-of 2026-09-14\nsupporting locator: “Time to live (TTL) configuration”; Memory Bank configuration examples\nevidence label: VERIFIED_SPEC + PRODUCT_VOLATILE\nrelation: support + limit\nfinding: TTL未設定ならcreated/updated memoryは自動expireしない。TTLはcreate/update/generate等のoperation別に設定でき、update時に新しいexpirationへ更新され得る。revisionとcurrent memoryにも別寿命が存在する。\n\n\n4) source class: official product documentation\nexact title: Enable agent memory\nauthor/org: Amazon Web Services\ncanonical URL: https://docs.aws.amazon.com/bedrock/latest/userguide/agents-configure-memory.html\npublication/update/access date: accessed 2026-09-14\nversion/release/API revision/model ID: current Amazon Bedrock Agents documentation as-of 2026-09-14\nsupporting locator: Memory duration / memoryConfiguration / storageDays / memoryId note\nevidence label: VERIFIED_SPEC + PRODUCT_VOLATILE\nrelation: support + availability gate\nfinding: Bedrock Agent memoryはSESSION_SUMMARY型、retention 1–365日、既定30日。memoryIdをinvoke時に指定しないturnはmemoryへ保存されない。model/Region availability gateあり。\n\n\n5) source class: official product documentation\nexact title: Create a memory store (AgentCore Memory)\nauthor/org: Amazon Web Services\ncanonical URL: https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/memory-create-a-memory-store.html\npublication/update/access date: accessed 2026-09-14\nversion/release/API revision/model ID: current AgentCore Memory documentation as-of 2026-09-14\nsupporting locator: event retention / eventExpiryDuration\n evidence label: VERIFIED_SPEC + PRODUCT_VOLATILE\nrelation: support + limit\nfinding: raw eventのshort-term memory expiryは最大365日。eventExpiryDurationはwrite時にeventごとへ適用され、設定変更は将来eventにのみ作用し、既存eventは元のexpirationを保持する。retention設定変更のretroactive effectを仮定できない。\n\n\n6) source class: standard/guidance\nexact title: NIST Privacy Framework: A Tool for Improving Privacy through Enterprise Risk Management, Version 1.0\nauthor/org: NIST\ncanonical URL: https://nvlpubs.nist.gov/nistpubs/CSWP/NIST.CSWP.01162020.pdf\npublication/update/access date: 2020-01-16 / accessed 2026-09-14\nversion/release: Version 1.0\nsupporting locator: data action/data processing definitions; CT.PO-P2; CT.PO-P4; CT.DM-P4/P5/P8/P9\nevidence label: VERIFIED_SPEC\nrelation: context + support\nfinding: data lifecycleをcollectionからdisposalまで扱い、retention/logging/generation/transformation/use/disclosure/transmission/disposalをdata actionに含める。削除・破壊・log・技術的対策のtest/assessmentまでpolicy対象とする基礎。\n\n\n7) source class: standard/guidance\nexact title: Guidelines for Media Sanitization\nauthor/org: Ramaswamy Chandramouli (NIST), Eric A. Hibbard (Samsung Semiconductor) / NIST\ncanonical URL: https://csrc.nist.gov/pubs/sp/800/88/r2/final\npublication/update/access date: 2025-09-26 final / accessed 2026-09-14\nversion/release: NIST SP 800-88 Rev.2\nsupporting locator: abstract; media sanitization program; distributed/cloud propagation; sanitization outcome/evidence guidance\nevidence label: VERIFIED_SPEC\nrelation: context + limit\nfinding: sanitizationはtarget dataへのaccessを一定労力でinfeasibleにする概念で、分散/cloudでは複数system/organizationへdataが伝播し得る。削除APIの成功をmedia-level sanitization証明へ外挿してはならず、期待結果とevidenceを定める必要がある。\n\n\n8) source class: primary research paper\nexact title: From Recall to Forgetting: Benchmarking Long-Term Memory for Personalized Agents\nauthor/org: Md Nayem Uddin, Kumar Shubham, Eduardo Blanco, Chitta Baral, Gengyu Wang / Findings of ACL 2026\ncanonical URL: https://aclanthology.org/2026.findings-acl.1337/\npublication/update/access date: July 2026 / accessed 2026-09-14\nversion/release: Findings ACL 2026; DOI 10.18653/v1/2026.findings-acl.1337\nsupporting locator: abstract; Forgetting-Aware Memory Accuracy (FAMA); benchmark comparison and evaluation sections\nevidence label: RESEARCH\nrelation: support + counter/limit\nfinding: Memoraはmemory mutation/deletion後のobsolete/invalid memory再利用を評価し、FAMAは有効memoryの利用だけでなく無効化・削除済みmemoryへの依存を罰する。複数memory agentでobsolete informationの再利用が観測された。ただしこれはbehavioral forgetting評価であり、storage purgeの証明ではない。" }, { "heading": "FINDINGS", "text": "- VERIFIED_SPEC: “memory”はprovider内でも単一artifactではない。OpenAIはabuse log/application state、Googleはcurrent Memory/revision、AWSはraw event/session summary等を別レイヤとして扱う。よってEnterprise AI Coreはartifact classごとにlifecycleを追う必要がある。\n- VERIFIED_SPEC: deleteは即時・全面消去と同義ではない。Google Memory Bankは親memory削除後もrevisionを最大48時間参照・rollback可能。OpenAIでも一部stateful objectは削除後server deletionまで30日とされる。\n- VERIFIED_SPEC: TTLの作用範囲と更新規則は製品固有。Googleではupdate/generateがexpirationを再計算し得る。AWS AgentCoreではretention設定変更が既存eventへ遡及しない。したがって「retentionを30日に変更したから既存dataも30日以内に消える」とは一般化できない。\n- INFERENCE: summarization/extraction/consolidationは派生data生成である。raw sessionを削除してもderived summary/fact/revision/indexが連鎖削除されるかは個別仕様または実測で確認すべきである。\n- PROPOSAL: Coreに `AI Memory Lifecycle Matrix / Artifact Deletion Graph` を置き、少なくとも artifact_type / source / derived_from / scope(user, session, agent, org) / store / create_time / update_time / TTL / retention_rule / delete_method / delete_requested_at / logical_inaccessibility_at / expected_purge_at / residual_window / external_copy / evidence を管理する。\n- PROPOSAL: “reset”は曖昧語として禁止せず、必ずdestructive scope manifestを伴わせる。reset対象にsession、durable memory、revision、vector/index、files、logs、tool/MCP downstream stateのどこまでを含むかを明示する。" }, { "heading": "LIMITS / COUNTEREVIDENCE / FAILURE ENVELOPE", "text": "- APIの404、listからの消失、agentが回答しなくなったことは、それぞれlogical deletion / retrieval absence / behavioral forgettingの証拠であり、物理媒体・backup・replica・派生物までの消去を単独では証明しない。\n- 逆にstorage objectを削除しても、同じfactが別session、summary、revision、RAG corpus、tool/MCP、human-authored profile等に残ればagentは再び同内容を取得し得る。\n- provider public docsはtenant固有legal hold、契約、safety exception、backup内部実装を完全には証明しない。\n- NIST SP 800-88r2はmedia sanitization基準であり、SaaS APIのdelete endpointにそのまま適用できるとの証拠ではない。ここでは「削除成功≠media sanitization」の境界根拠として使用。\n- Memora/FAMAはbehavioral memory correctnessを測るbenchmarkで、provider storage deletionやregulatory erasure保証へ外挿しない。\n- model-weight training/unlearningはapplication memory lifecycleとは別問題。本Packetからweight-level消去を推論しない。" }, { "heading": "AVAILABILITY_GATES", "text": "- OpenAI ZDR/MAMはeligible customer向け承認制で、endpoint/capabilityごとにZDR適格性が異なる。safety/legal例外もあり得る。\n- Google Memory Bankはproject/API/IAM/region等の要件を持つ。revision/TTL semanticsは現行Memory Bank固有であり他Google productへ一般化しない。\n- AWS Bedrock memoryは対応model/RegionとmemoryId等の設定に依存。AgentCore MemoryとBedrock Agents session-summary memoryを同一実装として扱わない。\n- organization retention、legal hold、regulated workload契約はprovider defaultより優先・追加され得るため、公開docだけでtenant実挙動を確定しない。" }, { "heading": "IMPLEMENTATION_STATE", "text": "現行主要providerにはTTL、delete/purge、revision、retention期間、ZDR等の個別primitiveが既に実装されている。一方、provider-neutralに「このresetで全派生memoryと残存surfaceが消えた」と表現・証明する共通schema/standardは本RUNで確認できていない。Core実装はprovider adapterが各artifactをLifecycle Matrixへ写像する方式が現実的。" }, { "heading": "BENCHMARK_OR_EVAL", "text": "RESEARCH: Memora/FAMAは、最新memory stateに対してobsolete/deleted informationを使わないかを測る有力なbehavioral forgetting評価である。\nPROPOSAL TEST_PROFILE（未実施・synthetic data only）:\n1. synthetic canary factをraw session/eventへ投入し、summary/long-term memory/revision/indexが生成されたIDと時刻を記録。\n2. TTL、retention、revision設定、provider/API revisionを記録。\n3. scoped delete/resetを実施し、即時にget/list/search/retrieve/semantic recallを確認。\n4. providerが明示するrecovery/purge window後に同一検査を再実施。\n5. exposedであればrevision、derived summary、vector/index、file、audit/log、tool/MCP downstream storeも別々に確認。\n6. paraphraseを含むnegative recall testでobsolete fact再利用を測る。ただしnegative recallはstorage erasure証明とはしない。\n7. deletion request ID、completion、expected residual window、各surfaceの結果をevidenceとして保存。" }, { "heading": "SECURITY/PRIVACY", "text": "長期memoryはpersonal/sensitive contextの保持期間とblast radiusを増やす。stale/obsolete memoryの再利用もprivacy・correctness riskになる。\nAUDIT: delete/resetはscope、resource ID、artifact class、provider/API revision、request/operation ID、request/completion time、expected residual window、verification結果を残す。監査log自体のretention/minimizationも別管理する。" }, { "heading": "COST", "text": "長期保存、revision、index、backup、定期purge、残存検証にはstorage/operation/evaluation costがある。保持しない設計はprivacyだけでなくcost controlにもなる。" }, { "heading": "OPERABILITY", "text": "asynchronous deletion、eventual consistency、recovery grace window、TTL refresh、non-retroactive config変更を運用runbookへ組み込む。reset完了通知を単一booleanに縮約しない。" }, { "heading": "SAFE_WORDING_FOR_FUTURE_BOOK", "text": "「Enterprise AIのmemory lifecycleは『覚える／忘れる』の二値ではない。session/raw event、派生summary、durable memory、revision、retrieval index、application state、logなどに別々のretentionとdelete semanticsが存在し得る。delete/resetを運用要件にする場合は、対象scopeとartifact、期待される残存期間、確認方法を分離して記録する。API上の404やagentの非想起だけから、全派生物・backup・物理媒体までの消去を断定しない。」" }, { "heading": "UPDATE_WORK_HANDOFF", "text": "- 将来本文ではEA-063 Privacy Engineeringと接続しつつ、privacy原則ではなく「実装・運用としての寿命管理」に独立節を置く。\n- provider mappingはOpenAI=endpoint別application-state/abuse-log/ZDR、Google=Memory/Revision/TTL/recovery window、AWS=SESSION_SUMMARY retentionとAgentCore raw-event expiryを別表にする。\n- Core用 `AI Memory Lifecycle Matrix / Artifact Deletion Graph` の採否を監査する。\n- reset/deletion手順には上記synthetic TEST_PROFILEを添え、認証tenantで必要な場合のみ後続Workが安全審査後に実施する。\n- 2026年以降のmemory APIは変動が速いため、公開後アップデート時には各providerのretention/delete/revision/ZDR文言を必ず再確認する。" }, { "heading": "OPEN_UNKNOWN", "text": "- provider横断で利用できるreset/forgetting completeness証明schemaは未確認。\n- backup/replica/low-level purgeの多くはpublic APIから直接観測不能で、契約・内部control evidenceが必要になり得る。\n- raw source削除がderived summary/embedding/indexへtransitive deleteされる条件はproduct別に未確定。\n- deletionと同時進行のmemory generation/consolidation、eventual consistency中のrace behaviorは未検証。\n- external MCP/tool/third-party memoryへの派生data削除は各第三者policy/API依存。" }, { "heading": "STOP_CONDITION", "text": "multi-artifact lifecycle、providerごとのretention/delete差、delete後残存window、behavioral forgettingとstorage deletionの分離、bounded residual TEST_PROFILEまで確立したため、本EAの先行採集は一旦停止。次回は未調査最小EAへ進む。\n\n---" }, { "heading": "AUDITED_BOUNDARY", "text": "Raw session/events, summaries, durable memory, revisions, vector/index state, files/application state, logs and third-party state can have independent lifecycle semantics. Delete/reset, TTL expiry, behavioral forgetting and storage sanitization are not synonyms. Negative recall is not storage-erasure proof. AWS Bedrock user-guide/API storageDays discrepancy remains explicit." } ], "human_url": "/tools/ea-064-ai-data-memory-lifecycle", "markdown_url": "/agent/article.md?path=%2Ftools%2Fea-064-ai-data-memory-lifecycle", "json_url": "/agent/article.json?path=%2Ftools%2Fea-064-ai-data-memory-lifecycle" }