{
  "format": "agent-article-v0",
  "article_id": "EA-238",
  "canonical_url": "/tools/ea-238-pii-redaction",
  "page_type": "reference",
  "claim_ids": [],
  "evidence_ids": [
    "EA-238-CURRENT-APPROVAL",
    "EA-238-CURRENT-BODY-READBACK",
    "EVD-EA-238-URL-423757B28269",
    "EVD-EA-238-URL-D3605CE106BF",
    "EVD-EA-238-URL-B518C9BF6F16",
    "EVD-EA-238-URL-A7254A51CF07",
    "EVD-EA-238-URL-A26C531A1FB1",
    "EVD-EA-238-URL-442F44134BE7",
    "EVD-EA-238-URL-B709DFD7C39D",
    "EVD-EA-238-URL-CA31AC8D1BD7"
  ],
  "unknown_ids": [
    "EA-238-UNKNOWN-01"
  ],
  "operational_object_ids": [],
  "provider": [],
  "models": [],
  "plans": [],
  "surfaces": [],
  "regions": [],
  "current_status": "CURRENT",
  "observation_status": "SPEC_ONLY",
  "freshness": "CURRENT",
  "last_verified": "2026-10-05",
  "supersedes": [],
  "superseded_by": [],
  "provenance": [
    "CANDIDATE-EA238-FIRST-20261005-C1",
    "BODY-SHA256:9e128c59e3eb1f8754488276a71fb0670b6e3cdbc83c813695853d2cd0e8ff3d",
    "SRC-EA-238-URL-423757B28269",
    "SRC-EA-238-URL-D3605CE106BF",
    "SRC-EA-238-URL-B518C9BF6F16",
    "SRC-EA-238-URL-A7254A51CF07",
    "SRC-EA-238-URL-A26C531A1FB1",
    "SRC-EA-238-URL-442F44134BE7",
    "SRC-EA-238-URL-B709DFD7C39D",
    "SRC-EA-238-URL-CA31AC8D1BD7"
  ],
  "article_sources": [
    {
      "id": "EA-238-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": [
        "CANDIDATE-EA238-FIRST-20261005-C1"
      ],
      "supporting_locators": [
        "1KqhJtiSb2S84rWmeYVwKcT_76dBwZ_B0ZZuAWQqH0rQ",
        "EA_REVISION_LEDGER!A41:U41",
        "EA-238-FIRST-20261005"
      ],
      "provenance_status": "CURRENT_CANONICAL_APPROVAL_READBACK"
    },
    {
      "id": "EA-238-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": "2026-10-12",
      "origin_class": "IMPLEMENTATION_READBACK",
      "source_or_experiment_ids": [
        "BODY-SHA256:9e128c59e3eb1f8754488276a71fb0670b6e3cdbc83c813695853d2cd0e8ff3d"
      ],
      "supporting_locators": [
        "content/staging/EA-238.md",
        "9e128c59e3eb1f8754488276a71fb0670b6e3cdbc83c813695853d2cd0e8ff3d",
        "EA-238-FIRST-20261005"
      ],
      "provenance_status": "CURRENT_APPROVED_BODY_HASH_MATCH"
    },
    {
      "id": "EVD-EA-238-URL-423757B28269",
      "evidence_class": "RESEARCH",
      "title": "1. Official current — Google Cloud, “Sensitive Data Protection documentation” / “InfoTypes and infoType detectors”, accessed 2026-10-05. Over 200 built-in detectors; custom detecto",
      "href": "https://docs.cloud.google.com/sensitive-data-protection/docs",
      "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-238-URL-423757B28269"
      ],
      "supporting_locators": [
        "Evidence table",
        "1. Official current — Google Cloud, “Sensitive Data Protection documentation” / “InfoTypes and infoType detectors”, accessed 2026-10-05. Over 200 built-in detectors; custom detectors; transformations; explicit warning that built-ins are not perfectly accurate and cannot guarantee regulatory compliance. https://docs.cloud.google.com/sensitive-data-protection/docs"
      ],
      "scope": "Evidence table",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-238-URL-D3605CE106BF",
      "evidence_class": "RESEARCH",
      "title": "1. Official current — Google Cloud, “Sensitive Data Protection documentation” / “InfoTypes and infoType detectors”, accessed 2026-10-05. Over 200 built-in detectors; custom detecto",
      "href": "https://docs.cloud.google.com/sensitive-data-protection/docs/concepts-infotypes",
      "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-238-URL-D3605CE106BF"
      ],
      "supporting_locators": [
        "Evidence table",
        "1. Official current — Google Cloud, “Sensitive Data Protection documentation” / “InfoTypes and infoType detectors”, accessed 2026-10-05. Over 200 built-in detectors; custom detectors; transformations; explicit warning that built-ins are not perfectly accurate and cannot guarantee regulatory compliance. https://docs.cloud.google.com/sensitive-data-protection/docs ; https://docs.cloud.google.com/sensitive-data-protection/docs/concepts-infotypes"
      ],
      "scope": "Evidence table",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-238-URL-B518C9BF6F16",
      "evidence_class": "RESEARCH",
      "title": "2. Official current — Google Cloud, “De-identification”, accessed 2026-10-05. Inspection configuration controls detector selection and transformations.",
      "href": "https://docs.cloud.google.com/sensitive-data-protection/docs/concepts-de-identification",
      "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-238-URL-B518C9BF6F16"
      ],
      "supporting_locators": [
        "Evidence table",
        "2. Official current — Google Cloud, “De-identification”, accessed 2026-10-05. Inspection configuration controls detector selection and transformations. https://docs.cloud.google.com/sensitive-data-protection/docs/concepts-de-identification"
      ],
      "scope": "Evidence table",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-238-URL-A7254A51CF07",
      "evidence_class": "RESEARCH",
      "title": "Amazon Comprehend”, accessed 2026-10-05. English/Spanish text PII; locating can be realtime/batch; redaction requires asynchronous batch.",
      "href": "https://docs.aws.amazon.com/comprehend/latest/dg/pii.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-238-URL-A7254A51CF07"
      ],
      "supporting_locators": [
        "Evidence table",
        "- Amazon Comprehend”, accessed 2026-10-05. English/Spanish text PII; locating can be realtime/batch; redaction requires asynchronous batch. https://docs.aws.amazon.com/comprehend/latest/dg/pii.html"
      ],
      "scope": "Evidence table",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-238-URL-A26C531A1FB1",
      "evidence_class": "RESEARCH",
      "title": "4. Standard/guidance — NIST SP 800-188, “De-Identifying Government Datasets: Techniques and Governance”, 2023. De-identification requires risk/governance analysis; masking alone ma",
      "href": "https://doi.org/10.6028/NIST.SP.800-188",
      "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-238-URL-A26C531A1FB1"
      ],
      "supporting_locators": [
        "Evidence table",
        "4. Standard/guidance — NIST SP 800-188, “De-Identifying Government Datasets: Techniques and Governance”, 2023. De-identification requires risk/governance analysis; masking alone may be insufficient; re-identification studies are relevant. https://doi.org/10.6028/NIST.SP.800-188"
      ],
      "scope": "Evidence table",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-238-URL-442F44134BE7",
      "evidence_class": "RESEARCH",
      "title": "5. Standard/guidance — NIST IR 8053, “De-Identification of Personal Information”, 2015. De-identification reduces privacy risk but some de-identified data can be re-identified.",
      "href": "https://doi.org/10.6028/NIST.IR.8053",
      "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-238-URL-442F44134BE7"
      ],
      "supporting_locators": [
        "Evidence table",
        "5. Standard/guidance — NIST IR 8053, “De-Identification of Personal Information”, 2015. De-identification reduces privacy risk but some de-identified data can be re-identified. https://doi.org/10.6028/NIST.IR.8053"
      ],
      "scope": "Evidence table",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-238-URL-B709DFD7C39D",
      "evidence_class": "RESEARCH",
      "title": "6. Author-maintained issue — data-privacy-stack/Presidio #1925, opened 2026-03-31, version 2.2.362 reproduction: anonymizer could leave later space-separated entities exposed; clos",
      "href": "https://github.com/data-privacy-stack/presidio/issues/1925",
      "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-238-URL-B709DFD7C39D"
      ],
      "supporting_locators": [
        "Evidence table",
        "6. Author-maintained issue — data-privacy-stack/Presidio #1925, opened 2026-03-31, version 2.2.362 reproduction: anonymizer could leave later space-separated entities exposed; closed via #1932. https://github.com/data-privacy-stack/presidio/issues/1925"
      ],
      "scope": "Evidence table",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-238-URL-CA31AC8D1BD7",
      "evidence_class": "RESEARCH",
      "title": "7. Author-maintained issue — Presidio #1845, opened 2026-02-04: plain hash anonymizer output could be brute-forced/dictionary-attacked; closed via #1846.",
      "href": "https://github.com/data-privacy-stack/presidio/issues/1845",
      "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-238-URL-CA31AC8D1BD7"
      ],
      "supporting_locators": [
        "Evidence table",
        "7. Author-maintained issue — Presidio #1845, opened 2026-02-04: plain hash anonymizer output could be brute-forced/dictionary-attacked; closed via #1846. https://github.com/data-privacy-stack/presidio/issues/1845"
      ],
      "scope": "Evidence table",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    }
  ],
  "research_sections": [
    {
      "heading": "要点と前提",
      "text": "PII検出器とredaction/anonymization処理は、入力から識別子を見つけて変換するための部品であり、それだけで「匿名化済み」「規制適合」「再識別不能」を保証するものではない。検出対象、言語、文脈、データ形式、しきい値、変換方式、復元鍵・対応表、ログやキャッシュなどの残存コピーを分けて設計・検証する。"
    },
    {
      "heading": "運用原則",
      "text": "1. **検出と匿名化を分離する。** detectorのfalse negativeは変換器では回復できない。Google Sensitive Data Protectionもbuilt-in infoType detectorを完全精度・規制適合保証として扱わず、要件に対する設定試験を求めている。\n2. **対象面を明示する。** text、structured data、file、image、stream等で利用可能な検出・変換経路は異なる。Amazon ComprehendのPII redactionはtext向けで、redactionは非同期batch jobである。製品名だけから全channel対応を推論しない。\n3. **言語・domainを固定して評価する。** 独立benchmarkではPII detectorの性能がdomain・script・entity typeで大きく変動する。単一英語fixtureのPASSを多言語・医療・法務・金融等へ外挿しない。\n4. **変換方式をprivacy claimと同一視しない。** masking/redaction/tokenization/hash/pseudonymizationは異なる性質を持つ。NISTはde-identification後にも再識別リスクが残り得るとしており、単なるmasking toolが十分なde-identification機能を持つとは限らない。\n5. **linkabilityとreversibilityを設計対象にする。** deterministic hashやstable tokenはjoin可能性を残す。Presidioではplain hash operatorの辞書・総当たり再識別リスクが2026年にissue化され、修正対象となった。\n6. **failure envelopeを試験する。** 連続entity、短いID、文脈依存の氏名・住所、数値の曖昧性、Unicode/非Latin、OCR由来text、chunk境界、重複・overlap entityを含める。Presidio 2.2.362ではspace-separated同種entityの一部がanonymizeされない再現issueが報告され、その後修正された。\n7. **残存コピーを別管理する。** 原文prompt、inspection findings、debug log、trace、cache、queue、backup、SIEM、analytics、human-review copyにPIIが残る可能性を、変換後payloadの確認とは別に監査する。"
    },
    {
      "heading": "Failure envelope / OPEN_UNKNOWN",
      "text": "対象組織のPII定義、言語・domain、入力channel、選定製品/version、custom detectors、precision/recall threshold、hash/token key管理、human-review経路、logging/retention、再識別リスク評価、法域別要件、live tenant挙動は未確定である。これらが未検証の状態では、公開claimは「PII検出・変換の部品と検証設計」に限定する。"
    },
    {
      "heading": "Evidence boundary",
      "text": "Google Sensitive Data Protectionの現行documentationは200超のbuilt-in infoTypes、custom detectors、複数のde-identification方式を示す一方、built-in detectorが完全精度ではなく規制適合を保証しないと明記する。Amazon Comprehendは英語/スペイン語textのPII検出とredactionを提供するが、redactionはasync batchである。NIST SP 800-188 / IR 8053はde-identificationをrisk-reductionとして扱い、再識別可能性とgovernanceを要求する。2026年のPIIBench等はdomain横断で検出性能の不均一性を示す。これらは特定tenantの品質保証ではない。"
    }
  ],
  "human_url": "/tools/ea-238-pii-redaction",
  "markdown_url": "/agent/article.md?path=%2Ftools%2Fea-238-pii-redaction",
  "json_url": "/agent/article.json?path=%2Ftools%2Fea-238-pii-redaction"
}