<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom" xml:lang="en">
  <title>VAP Blog — Verifiable AI Provenance Framework</title>
  <link href="https://vap.veritaschain.org/blog/"/><link rel="self" href="https://vap.veritaschain.org/feed.xml"/>
  <id>https://vap.veritaschain.org/</id>
  <updated>2026-08-19T00:00:00+09:00</updated>
  <author><name>VeritasChain Standards Organization</name></author>
  <entry>
    <title>One Photograph of a Screen: FACES and Investigative Decision Records</title>
    <link href="https://vap.veritaschain.org/blog/dillon-faces-investigative-records/"/>
    <id>https://vap.veritaschain.org/blog/dillon-faces-investigative-records/</id>
    <updated>2026-08-19T00:00:00+09:00</updated>
    <summary>A 93% FACES match, a phone photo of a screen, and a June 2026 federal suit — why investigative decisions that never became records motivated VAP's PAP vocabulary.</summary>
  </entry>
  <entry>
    <title>40 of 45: The KPMG Citation Failure and the Limits of Provenance</title>
    <link href="https://vap.veritaschain.org/blog/kpmg-citations-provenance-limits/"/>
    <id>https://vap.veritaschain.org/blog/kpmg-citations-provenance-limits/</id>
    <updated>2026-08-19T00:00:00+09:00</updated>
    <summary>GPTZero found 40 of 45 citation titles fabricated in a withdrawn KPMG report. Provenance frameworks record origin and integrity — they do not verify truth.</summary>
  </entry>
  <entry>
    <title>17,000 Events Later: The Hugging Face Breach and Log Completeness</title>
    <link href="https://vap.veritaschain.org/blog/hugging-face-log-completeness/"/>
    <id>https://vap.veritaschain.org/blog/hugging-face-log-completeness/</id>
    <updated>2026-08-19T00:00:00+09:00</updated>
    <summary>Hugging Face reconstructed an attacker log of 17,000+ events. The early 'incomplete logs' story was wrong — and that is the point: completeness must be verified.</summary>
  </entry>
  <entry>
    <title>Doe v. Meta: When the Selection Records Are the Dispute</title>
    <link href="https://vap.veritaschain.org/blog/doe-v-meta-selection-records/"/>
    <id>https://vap.veritaschain.org/blog/doe-v-meta-selection-records/</id>
    <updated>2026-08-19T00:00:00+09:00</updated>
    <summary>Twenty-six plaintiffs say AI systems shaped Meta's May 2026 layoffs; Meta says people decided. The dispute sits exactly on VAP's acknowledged decision-authority gap.</summary>
  </entry>
  <entry>
    <title>Eto v. Mayo Clinic: Who Reviewed the De-Identification?</title>
    <link href="https://vap.veritaschain.org/blog/eto-v-mayo-oversight-records/"/>
    <id>https://vap.veritaschain.org/blog/eto-v-mayo-oversight-records/</id>
    <updated>2026-08-19T00:00:00+09:00</updated>
    <summary>A director who supported Mayo's IRB alleges AI-era de-identification went unreviewed — untested claims that show why oversight roles need tamper-evident records.</summary>
  </entry>
  <entry xml:lang="ja">
    <title>画面を撮った一枚の写真 — FACES誤認逮捕と捜査判断の記録</title>
    <link href="https://vap.veritaschain.org/ja/blog/dillon-faces-investigative-records/"/>
    <id>https://vap.veritaschain.org/ja/blog/dillon-faces-investigative-records/</id>
    <updated>2026-08-19T00:00:00+09:00</updated>
    <summary>FACESの「93%一致」、画面を携帯で撮った一枚、2026年6月の連邦提訴——記録にならなかった捜査判断が、PAP捜査判断語彙の起草を動機づけた。</summary>
  </entry>
  <entry xml:lang="ja">
    <title>45件中40件 — KPMG引用崩壊と、プロビナンスが検証できないもの</title>
    <link href="https://vap.veritaschain.org/ja/blog/kpmg-citations-provenance-limits/"/>
    <id>https://vap.veritaschain.org/ja/blog/kpmg-citations-provenance-limits/</id>
    <updated>2026-08-19T00:00:00+09:00</updated>
    <summary>撤回されたKPMGレポートで、45件中40件の引用タイトルが捏造とGPTZeroが判定。プロビナンス枠組みが記録するのは出所と完全性——真実の検証ではない。</summary>
  </entry>
  <entry xml:lang="ja">
    <title>17,000イベントの後で — Hugging Face侵害とログ完全性</title>
    <link href="https://vap.veritaschain.org/ja/blog/hugging-face-log-completeness/"/>
    <id>https://vap.veritaschain.org/ja/blog/hugging-face-log-completeness/</id>
    <updated>2026-08-19T00:00:00+09:00</updated>
    <summary>Hugging Faceは17,000件超の攻撃者行動ログを再構成した。初期の「不完全なログ」説は誤りだった——完全性は主張ではなく検証の対象である、というのが教訓だ。</summary>
  </entry>
  <entry xml:lang="ja">
    <title>Doe対Meta — 選別の記録こそが争点になるとき</title>
    <link href="https://vap.veritaschain.org/ja/blog/doe-v-meta-selection-records/"/>
    <id>https://vap.veritaschain.org/ja/blog/doe-v-meta-selection-records/</id>
    <updated>2026-08-19T00:00:00+09:00</updated>
    <summary>26名の原告はAIシステムが2026年5月のレイオフ選別を形作ったと主張し、Metaは「決めたのは人」と否認。争点はVAPが自認する意思決定権限配分ギャップの真上にある。</summary>
  </entry>
  <entry xml:lang="ja">
    <title>Eto対Mayo Clinic — 非識別化を誰がレビューしたのか</title>
    <link href="https://vap.veritaschain.org/ja/blog/eto-v-mayo-oversight-records/"/>
    <id>https://vap.veritaschain.org/ja/blog/eto-v-mayo-oversight-records/</id>
    <updated>2026-08-19T00:00:00+09:00</updated>
    <summary>MayoのIRBを支えた研究運営ディレクターが、AI時代の非識別化が未レビューだったと主張（未検証）。監督の役割に改ざん検知可能な記録が要る理由を示す事案。</summary>
  </entry>
  <entry>
    <title>The Records That Were Never Created</title>
    <link href="https://vap.veritaschain.org/blog/records-never-created/"/>
    <id>https://vap.veritaschain.org/blog/records-never-created/</id>
    <updated>2026-08-19T00:00:00+09:00</updated>
    <summary>Why the hardest problem in AI oversight is not detecting altered records but detecting absent ones — and how VAP's three-tier taxonomy and Completeness Invariant (INT-008) address the part that can be addressed.</summary>
  </entry>
  <entry xml:lang="ja">
    <title>作られなかった記録</title>
    <link href="https://vap.veritaschain.org/ja/blog/records-never-created/"/>
    <id>https://vap.veritaschain.org/ja/blog/records-never-created/</id>
    <updated>2026-08-19T00:00:00+09:00</updated>
    <summary>AI監督の最難関は、改ざんされた記録の検知ではなく、存在しない記録の検知である — VAPの三層分類と完全性インバリアント（INT-008）が、対処可能な部分にどう対処するか。</summary>
  </entry>
</feed>
