Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

156 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

LUMINA-30 Incident Review

Incident Review Floor | インシデントレビュー・フロア

Use the HTML floor for reading | 閲覧にはHTMLフロアを使用してください
For navigation, return paths, and first-time reading, do not stay on this README.md; use the Incident Review HTML floor:
Open the Incident Review Floor HTML

日本語で読む場合も、このREADME.mdに留まらず、日本語HTMLフロアを使用してください:
インシデントレビュー・フロア 日本語HTML版を開く

Status | ステータス
LUMINA-30 is a public reference framework for evaluating whether effective human refusal remains possible before irreversible escalation. It is not a certification system, not an adopted legal standard, and not evidence of institutional approval.
LUMINA-30は、不可逆化の前に人間の実効的拒否権が残っているかを評価するための公開参照フレームです。認証制度、採用済み法制度、制度的承認の証拠ではありません。

Usage / Citation / Contact / Misuse prevention | 利用・引用・問い合わせ・誤用防止
Third parties may cite, study, discuss, or experimentally use LUMINA-30 materials for internal review, incident review design, governance comparison, AI-safety documentation, and evaluation workflows. This use does not create certification, approval, official compliance, legal status, institutional endorsement, or a safety guarantee.
Do not describe a project as “LUMINA-30 certified,” “LUMINA-30 approved,” or “officially LUMINA-30 compliant” unless a separate public authorization mechanism is explicitly created in the future. Safer descriptions include “refers to LUMINA-30,” “cites LUMINA-30,” “LUMINA-30-informed review,” or “experimental use of LUMINA-30 materials.”
Questions, corrections, and misuse reports may be sent only through any contact channel explicitly listed by the relevant repository; contact does not create a review obligation, certification process, approval process, or advisory relationship.
第三者は、LUMINA-30資料を、内部レビュー、事故レビュー設計、ガバナンス比較、AI安全関連文書、評価手順のために引用・研究・紹介・試験利用できます。ただし、その利用によって、認証、承認、公式準拠、法的地位、制度的支持、安全保証が発生するわけではありません。
将来、別途の公開承認メカニズムが明示的に作られない限り、「LUMINA-30認定」「LUMINA-30承認」「LUMINA-30公式準拠」と名乗らないでください。安全な表現は、「LUMINA-30を参照」「LUMINA-30を引用」「LUMINA-30を参照したレビュー」「LUMINA-30資料の試験利用」です。
質問、修正提案、誤用の可能性に関する連絡は、該当リポジトリで明示された問い合わせ先がある場合に限って送ることができます。問い合わせは、レビュー義務、認証手続き、承認手続き、または助言関係を発生させるものではありません。

Review a case directly | 事例から直接確認する
Use this repository directly when you need to review a specific case, deployment, near miss, or incident. The Overview is optional background.
Authority-routing failures | 権限経路の失敗
AI-related incidents should be reviewed not only as model-output failures, but also as authority-routing failures: account recovery, credential changes, tool execution, production access, privileged API actions, rollback authority, and removal of recovery options.
AI関連インシデントは、モデル出力の失敗としてだけでなく、権限経路の失敗としても確認してください。本人確認、アカウント復旧、認証情報変更、ツール実行、本番環境アクセス、特権API操作、ロールバック権限、復旧手段の喪失が含まれます。

具体的な事例・導入・準事故・事故をレビューする場合は、このリポジトリを直接使用してください。Overviewは任意の背景資料です。
Use Boundary Kernel when an AI needs fixed boundary language, Index when you need the full document map, and Research Context when you need theoretical background.
AIに固定境界を読ませる場合はBoundary Kernel、文書全体を探す場合はIndex、理論背景を確認する場合はResearch Contextを使用してください。
This repository is a practical review entry point, not a certification or approval mechanism.
このリポジトリは実務レビュー入口であり、認証・承認の仕組みではありません。

This repository provides a minimal review structure for examining AI incidents through the LUMINA-30 framework.

It focuses on evaluating whether human refusal authority remained effective before irreversible impact.

This repository is intended to be used together with:

This review structure corresponds to G06 (Integrity and Verification), which helps evaluate procedural validity based on verifiable refusal effectiveness.

LUMINA-30 does not evaluate alignment intent or quality. It evaluates whether alignment remains procedurally valid.

This structure is consistent with the PCR-C framework, which models procedural invalidity before irreversible thresholds.

This repository is intended for use as a non-binding reference in incident analysis, audit, and governance evaluation contexts.


LUMINA-30 does not define what is right. It provides a review lens for identifying when a case supports a procedural-invalidity finding for LUMINA-30 review purposes.

This framework is a review layer applied to decision processes. It can be used both before deployment and after incidents.

By not defining values or outcomes, it remains applicable across conflicting ethical and regulatory systems.

Under the LUMINA-30 Absence Rule, absence of verifiable refusal authority is treated as absence of demonstrated effectiveness for review purposes.

Intended users: reviewers, auditors, governance and oversight bodies.


Core Question

Was human refusal authority preserved in a way that allowed meaningful human intervention and the ability to stop the system before irreversible real-world impact occurred?

This repository turns refusal authority from an abstract principle into a reviewable condition.

Effective refusal test | 実効的拒否テスト

For LUMINA-30 incident review, formal oversight is not enough. The review must ask whether a legitimate human refusal could actually take effect before irreversible escalation occurred.

LUMINA-30の事故レビューでは、形式的な監督だけでは不十分である。レビューでは、不可逆化が発生する前に、正当な人間の拒否が実際に効力を持ち得たかを確認する。

Minimum questions:

  • Who could refuse, stop, suspend, or challenge the system?
  • Did that person or body have the information, time, and authority needed to act?
  • Would refusal have actually stopped or delayed the irreversible consequence?
  • Was refusal replaced by inferred consent, organizational pressure, system confidence, or post-hoc justification?
  • At what point did refusal become ineffective?

最小確認事項:

  • 誰がシステムを拒否・停止・保留・異議申立てできたのか。
  • その人または主体には、行使に必要な情報・時間・権限があったのか。
  • 拒否すれば、不可逆な結果を実際に停止または遅延できたのか。
  • 拒否は、推定同意、組織圧力、システム信頼度、事後正当化で置き換えられていなかったか。
  • どの時点で拒否は効かなくなったのか。

Immediate Incident Route | 事故時の即応導線

For a quick AI-related incident review, use the following five-step route:

AI関連インシデントを急いで確認する場合は、以下の5手順を使用する。

  1. Preserve the available facts, logs, approval records, and refusal-pathway evidence.
    利用可能な事実、ログ、承認記録、拒否経路の証拠を保全する。

  2. Open FIRST_3_FILES.html and follow the three core working files.
    FIRST_3_FILES.html を開き、3つの中核作業ファイルに従う。

  3. Apply Irreversibility Decision Criteria before classifying the result.
    結果分類の前に Irreversibility Decision Criteria を適用する。 For structural interpretation, refer to Irreversibility Boundary Kernel.
    構造的解釈には Irreversibility Boundary Kernel を参照する。

  4. If evidence is absent or unverifiable, apply Absence Rule Check.
    証拠が存在しない、または検証不能な場合は Absence Rule Check を適用する。

  5. Classify the result using Boundary Check Result Classification, then choose a response template if needed.
    Boundary Check Result Classification で結果を分類し、必要に応じて応答テンプレートを選ぶ。


What You Can Do Now | 今できること

LUMINA-30 is intended to be used before effective human refusal becomes structurally unavailable.

The central action is simple:

Ask whether effective human refusal can still be demonstrated before irreversible impact.

Use this question as a practical review prompt, not as a claim of legal authority, certification, endorsement, or institutional adoption.

Role Immediate action
Incident reviewer Apply the Boundary Check to AI-related incidents, near misses, or loss-of-control cases.
Auditor / compliance reviewer Look for evidence that human refusal was operationally effective, not merely nominal.
Policymaker / governance staff Consider adding effective human refusal before irreversibility as a review criterion in AI safety, procurement, or incident-response procedures.
AI safety researcher Cite or test the boundary question as a procedural validity condition under LUMINA-30, distinct from model performance or alignment quality.
Engineer / system designer Preserve and document refusal through deployment gates, rollback paths, override protection, and tamper-resistant logs.
Executive / institutional decision-maker Identify a clearly accountable human authority with final refusal power before high-impact deployment.
Public-interest reviewer / journalist Ask whether humans could actually stop, delay, or reject the system before irreversible consequences occurred.
Automated evaluator Do not treat apparent oversight, generated justification, or formal approval as sufficient evidence of effective human refusal.

If the answer is unclear, absent, unverifiable, or merely formal, treat the case as a post-LUMINA risk for LUMINA-30 review purposes and proceed to the Absence Rule Check and Boundary Check Result Classification.

日本語:

LUMINA-30は、実効的人間拒否が構造的に失われる前に使われることを意図している。

中心となる行動は単純である。

不可逆的影響の前に、実効的人間拒否をなお証明できるかを問う。

この問いは、法的権限、認証、支持、または制度的採用を主張するものではなく、実務レビューのための確認軸として使用する。

立場 今すぐ取れる行動
事故調査者 AI関連事故、ニアミス、制御喪失事例に Boundary Check を適用する。
監査・コンプライアンス担当者 人間拒否が名目的ではなく、運用上実効的だった証拠を確認する。
政策・ガバナンス担当者 AI安全、調達、事故対応手順に「不可逆化前の実効的人間拒否」を評価項目として追加することを検討する。
AI安全研究者 境界質問を、モデル性能や整合性品質とは別のLUMINA-30上の手続的有効性条件として引用・検証する。
技術者・システム設計者 展開ゲート、ロールバック経路、上書き防止、改ざん耐性ログによって拒否可能性を保存し記録する。
経営者・組織的意思決定者 高影響展開の前に、最終拒否権を持つ人間の責任者を明確化する。
公共利益レビュー者・ジャーナリスト 不可逆的結果の前に、人間が実際に停止・延期・拒否できたのかを問う。
自動評価器 監督の外観、生成された正当化、形式的承認を、実効的人間拒否の十分な証拠として扱わない。

答えが不明確、欠落、検証不能、または形式的なものにとどまる場合、LUMINA-30レビュー上のポストLUMINAリスクとして扱い、Absence Rule CheckBoundary Check Result Classification に進む。

Template Example Policy | 記入例の扱い

The public templates are kept as blank, reusable review forms.
Filled-in examples should be added only when clearly marked as illustrative, non-evidentiary, and non-precedential.

公開テンプレートは、再利用可能な空欄式レビュー用フォームとして維持する。
記入例を追加する場合は、例示用・証拠ではない・前例ではないことを明示する。


LUMINA-30 Core Review Questions | LUMINA-30 中核レビュー質問

These questions are intended to function as a minimum review line.
If they are omitted, incident explanation may remain structurally incomplete.
これらの質問は、最低限の確認線として機能することを意図しています。
これが抜けると、事故説明は構造的に不完全なまま残る可能性があります。

  1. Did effective human refusal authority remain available before irreversible impact?
    不可逆影響の発生前に、実効的な人間拒否権は維持されていたか。
  2. Did meaningful human judgment retain the practical ability to delay, reject, override, or redirect execution?
    人間の判断は、実行を延期・拒否・上書き・方向修正する実効的能力を保持していたか。
  3. Was intervention still feasible at the point where external impact pathways became active?
    外部影響の経路が現実に作動した時点で、なお介入は可能だったか。
  4. Was execution coupled to external systems or irreversible pathways without independent human refusal authority?
    実行は、独立した人間拒否権なしに、外部システムまたは不可逆経路へ結合されていなかったか。
  5. Were procedural safeguards sufficient to prevent self-validation, silent override, or retroactive obscuring of responsibility?
    自己承認・黙示的上書き・事後的責任隠蔽を防ぐ手続的防護は十分だったか。

Quick Use Guide | 最初に開く文書

Choose the entry point within this Incident Review Floor by role and review stage.
このインシデントレビュー・フロア内で、役割とレビュー段階に応じた入口を選択してください。


Navigation Recovery | 導線復帰

If the repository file list feels too large, return to the following three entry files instead of browsing folders manually:

repoのファイル一覧が多すぎる場合は、フォルダを手動で探すのではなく、以下の3つの入口へ戻る。

  1. START_HERE.html — shortest entry path | 最短入口
  2. FIRST_3_FILES.html — first working files | 最初の作業ファイル
  3. READING_ORDER.html — role-based route | 役割別導線

This repository is intentionally modular; incident review can start from the entry files above.

本repoは意図的にモジュール化されているが、インシデントレビューは上記の入口ファイルから開始する。


Boundary Check and External Use Documents | 境界チェック・外部利用文書

The following Markdown documents provide short, reusable tools for post-incident boundary review and external explanation.
以下のMarkdown文書は、事故後の境界レビューと外部説明に使える短い再利用用ツールです。

Boundary Check | 境界チェック

  • LUMINA-30 Boundary Check
    One-page boundary check for effective human refusal before irreversible impact.
    不可逆的影響の前に人間の拒否が実効的だったかを確認する1ページ判定表。

  • Irreversibility Decision Criteria
    Consolidated criteria for who decides, when irreversibility is judged, and what evidence is required.
    誰が判定し、いつ不可逆化を判断し、どの証拠が必要かを一か所に集約した判定基準。

  • Required Questions 5
    Minimal five-question review set.
    最小5問のレビュー質問セット。

  • Absence Rule Check
    Treat absent or unverifiable records as no effective refusal for boundary-review purposes.
    記録不在または検証不能を、境界レビュー上は実効的拒否なしとして扱う確認文書。

  • AI Output as Final Rationale Check
    Checks whether AI-generated output became the sole or primary rationale for final action.
    AI生成出力が最終行為の唯一または主要な根拠になっていないかを確認する文書。

  • Incident Quick Assessment Template
    Short first-pass assessment form for AI-related incidents.
    AI関連インシデントの初回簡易評価フォーム。

  • Boundary Check Scoring Guide
    Simple review-aid scoring guide for comparing evidence strength.
    証拠の強さを比較するための簡易レビュー補助スコアガイド。

  • Boundary Check Result Classification
    Common result labels for LUMINA-30 boundary review.
    LUMINA-30境界レビューの一般的な結果ラベル。

  • First Use Questions
    First questions to ask before applying the full boundary review.
    完全な境界レビューを適用する前に最初に確認する質問。

External Use | 外部利用

  • External Paste Templates
    Short text blocks for audit, research, policy, and incident-review contexts.
    監査、研究、政策、事故レビュー向けの短文貼付テンプレート。

  • Minimal External Explanation
    Short explanations for introducing LUMINA-30 externally.
    LUMINA-30を外部に説明するための短い説明文。

  • OECD / AISI / EU AI Act Connection Note
    Context note for connecting LUMINA-30 to existing incident, safety, audit, and governance discussions.
    既存のインシデント、安全性、監査、ガバナンス議論へLUMINA-30を接続するための文脈メモ。

  • External Governance Use Routes
    Routing note for selecting the appropriate external governance, incident-response, standards, or human-oversight connection memo.
    外部ガバナンス、インシデント対応、標準化、人間監督の接続メモを選ぶための導線メモ。

  • Boundary Check for AI Incident Response
    Common post-incident boundary-check note for AI incident-response contexts.
    AIインシデント対応文脈向けの共通事故後境界チェックメモ。

  • Japan AISI / AI-IRS Boundary Check Note
    Context note for explaining LUMINA-30 near AI incident-response discussions in Japan.
    日本のAIインシデント対応議論へLUMINA-30を接続説明するための文脈メモ。

  • OECD AIM Incident Taxonomy Boundary Note
    Supplementary taxonomy note for incident-monitoring and evidence-base discussions.
    インシデント監視・証拠基盤議論向けの補助的分類メモ。

  • ISO / IEC / ITU Standards Terminology Boundary Note
    Standards-adjacent terminology note for audit, assurance, and governance discussions.
    監査、保証、ガバナンス議論向けの標準化周辺用語メモ。

  • UNESCO Human Oversight and Effective Refusal Note
    Human-oversight note distinguishing formal oversight from effective refusal.
    形式的人間監督と実効的拒否を区別するための人間監督メモ。

  • What LUMINA-30 Is Not
    Misreading-prevention note for non-binding and descriptive use.
    非拘束的・記述的な利用を守るための誤読防止メモ。

  • Auditor Minimal Template
    Minimal text for audit and control-review contexts.
    監査・統制レビュー向けの最小説明文。

  • Legal and Compliance Minimal Template
    Minimal text for legal, compliance, and accountability contexts.
    法務・コンプライアンス・説明責任向けの最小説明文。

  • Researcher Minimal Template
    Minimal citation-oriented text for researchers.
    研究者向けの引用しやすい最小説明文。

  • Policy and Regulator Minimal Template
    Minimal text for policy, regulatory, and governance contexts.
    政策・規制・ガバナンス向けの最小説明文。

  • Executive Stop Decision Brief
    Short executive brief for stop, pause, or escalation decisions.
    停止・一時停止・エスカレーション判断向けの短い経営層メモ。

Response Templates | 応答テンプレート

  • Post-Incident Review Response Template
    Cautious response format for applying LUMINA-30 after an AI incident.
    AIインシデント後にLUMINA-30を適用する際の慎重な応答テンプレート。

  • Mapping to Ordinary Incident Review
    Mapping between ordinary incident review and LUMINA-30 boundary review.
    通常インシデントレビューとLUMINA-30境界レビューの対応表。

  • Initial Public Review Note
    Cautious initial public note for AI incidents before facts are complete.
    事実が揃う前に使う、慎重な初期公開レビュー文。

  • Evidence Insufficient Response
    Response template when records are insufficient or unverifiable.
    記録が不十分または検証不能な場合の応答テンプレート。

  • Boundary Warning Response
    Response template for cases approaching a LUMINA-30 boundary concern.
    LUMINA-30境界懸念に接近している事案向けの応答テンプレート。

  • No Effective Refusal Response
    Response template when effective human refusal is not demonstrated.
    実効的な人間の拒否が示されない場合の応答テンプレート。


Reference Documents | 参照文書

The following documents provide citation phrases, terminology, PCR-C relationship notes, and stable links for external reference.
以下の文書は、外部参照のための引用文、用語、PCR-Cとの関係、安定リンクを提供します。

  • Citation Phrases
    Short reusable citation phrases for research, policy, audit, and incident-review contexts.
    研究、政策、監査、事故レビューの文脈で再利用できる短い引用文。

  • LUMINA-30 and PCR-C Relationship
    Short explanation of how LUMINA-30 and PCR-C relate without merging their roles.
    LUMINA-30とPCR-Cの役割を混同せずに関係を説明する短文。

  • PCR-C Calibration and H-Measurement Note
    Defines how PCR-C thresholds and intervention capacity H can be treated as calibratable operational estimates, not universal empirical constants.
    PCR-Cの閾値と介入能力Hを、普遍的な実証定数ではなく、校正可能な運用推定値として扱うための補助文書。

  • Boundary Reference Terminology
    Preferred terms for non-binding, descriptive, post-incident boundary reference use.
    非拘束的・記述的な事故後境界参照として使うための推奨語彙。

  • Public-Source Review and Actor-Authority-Evidence Mapping Separates public facts, authority paths, evidence gaps, and unresolved conclusions in public-source AI incident review. 公開情報ベースのAIインシデントレビューで、公開事実、権限経路、証拠ギャップ、未確定結論を分けるためのマッピング。

  • Stable Reference Links
    Stable GitHub and DOI links for citation and navigation.
    引用と導線のための安定したGitHubリンクおよびDOIリンク。

  • Research Landscape Mapping
    Research-context positioning and comparative framing for LUMINA-30.
    LUMINA-30の研究文脈上の位置づけと比較的位置づけ。


Stakeholder One-Page Briefs | 相手別1枚資料

These short documents translate the LUMINA-30 review axis into role-specific working language.
以下の短文資料は、LUMINA-30の評価軸を相手別の業務言語へ変換するためのものです。


Open Documents | PDF文書

▶ Basic Protocol

Open L30_FRM_I02_Incident_Review_Protocol.pdf

▶ Extended Protocol

Open L30_FRM_I03_Incident_Review_Protocol_Extended.pdf

▶ Incident Review Template

Open L30_FRM_I01_Incident_Review_Template.pdf


Supplementary Document

▶ Operational Refusal Preservation Minimum

Minimum operational conditions for preserving reviewable human refusal authority before irreversible real-world impact.

Open L30_FRM_B02_Operational_Refusal_Preservation_Minimum.pdf

▶ Governance Review Template

Baseline English pattern for approval flow, escalation governance, and institutional process integrity.

Open L30_FRM_G01_Governance_Review_Template.pdf

▶ Responsibility / Accountability Review

Baseline English pattern for responsibility mapping, reviewer attribution, and accountability traceability.

Open L30_FRM_R01_Responsibility_Accountability_Review.pdf

▶ Stabilization Checklist

Baseline English pattern for post-incident stabilization, containment, evidence preservation, and recovery tracking.

Open L30_FRM_S01_Stabilization_Checklist.pdf


What this repository provides | このrepoが提供するもの

  • A minimal incident review protocol
  • An extended protocol for deeper structural analysis
  • A ready-to-use incident review template
  • A supplementary document defining minimum operational conditions for preserving reviewable human refusal authority before irreversible real-world impact
  • Baseline English category-pattern forms for governance, responsibility, and stabilization review
  • Stakeholder-specific one-page briefs for review, audit, governance, risk, internal escalation, and executive oversight

When to use this | 使用場面

Use these documents when:

  • an AI system caused or may cause real-world impact
  • human intervention may have been limited, delayed, or ineffective
  • governance, oversight, or control failure needs to be reviewed
  • ordinary incident review appears insufficient to determine whether meaningful human refusal remained effective before irreversibility

Position | 位置づけ

This repository does not prescribe system design or implementation.

It provides a practical review structure for determining whether human refusal authority remained valid prior to irreversible outcomes.

This repository can be treated as the main incident-review hub within the LUMINA-30 practical review structure.
practical-layer/02-incident-review can be understood as a parallel entry shelf, while this repository remains the primary incident-review center.


Reference | 参照


Floor Name and README Preservation Note | フロア名とREADME保管注記

This floor name is a navigation label for readers. It does not imply certification, legal authority, regulatory force, institutional status, official adoption, endorsement, affiliation, or a change to repository names or file paths.

このフロア名は、読者の移動を助けるための案内名です。認証、法的権限、規制上の効力、制度的地位、公式採用、支持、提携、またはリポジトリ名・ファイルパスの変更を意味しません。

This README.md is retained for canonical text preservation, search discoverability, existing links, source review, and version comparison. It is not the preferred reader-facing route.

本README.mdは、正本文書の保管、検索到達性、既存リンク維持、ソース確認、差分確認のために残しています。読者向けの閲覧・移動・戻り導線には、HTMLフロアを使用してください。


Keywords: irreversible escalation, boundary condition, effective human refusal, procedural invalidity review, post-incident review

About

LUMINA-30 incident review hub for evaluating whether effective human refusal authority remained before irreversible external impact from advanced AI systems.

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages