Skip to content

Latest commit

 

History

History
169 lines (100 loc) · 9.55 KB

File metadata and controls

169 lines (100 loc) · 9.55 KB

LUMINA-30 Boundary Check | LUMINA-30 境界チェック

Purpose | 目的

This document provides a short, non-binding post-incident review check for examining whether meaningful human refusal authority remained effective before irreversible AI impact.

本書は、AIインシデント後レビューにおいて、不可逆的なAI影響の前に人間の拒否権が実効的に残っていたかを確認するための、短い非拘束的チェックである。

Status | 位置づけ

  • Non-binding reference

  • Descriptive review aid

  • Boundary check, not a legal conclusion

  • Post-incident review support

  • 非拘束的参照

  • 記述的なレビュー補助

  • 法的結論ではなく境界チェック

  • 事故後レビュー支援

Status Clarification | 位置づけの補足

Non-binding status limits institutional claims.
It does not weaken the boundary logic.

A review does not have to adopt LUMINA-30 as an official framework.
However, if the review concerns irreversible AI-related impact, it should still be able to show equivalent evidence that human refusal was available, operationally effective, and verifiable before irreversibility.

非拘束という位置づけは、制度的主張を制限するためのものである。
境界論理そのものを弱めるものではない。

レビューがLUMINA-30を公式フレームワークとして採用する必要はない。
しかし、そのレビューが不可逆的なAI関連影響を扱う場合、不可逆化前に人間の拒否が利用可能で、運用上実効的であり、検証可能だったことを、同等の証拠で示せる必要がある。

Terminology Note | 用語注記

This document uses operational boundary language.
Where the canonical framework uses “civilizational boundary,” this check reads it as a pre-irreversibility human-refusal boundary.

The question is not whether an event is culturally or historically “civilizational.”
The question is whether humans still had effective refusal authority before irreversible AI-related impact.

本書では、運用上の境界語を用いる。
正典フレームワークが「文明境界」と呼ぶものを、本チェックでは「不可逆化前の人間拒否境界」として読む。

問題は、ある出来事が文化的・歴史的な意味で「文明的」かどうかではない。
問題は、不可逆的なAI関連影響の前に、人間がなお実効的な拒否権を持っていたかである。

Core Question | 中核問い

Was human final refusal effective before irreversible impact?

不可逆的影響の前に、人間の最終拒否権は実効的だったか?

Boundary Check | 境界チェック

Answer each item as:

  • Yes
  • No
  • Not Verifiable

各項目は以下で回答する。

  • Yes
  • No
  • Not Verifiable

1. Human Final Refusal | 人間の最終拒否

Was human final refusal available before irreversible impact?

不可逆的影響の前に、人間の最終拒否は可能だったか。

2. Operational Effectiveness | 運用上の実効性

Was that refusal operationally effective, not merely nominal?

その拒否は名目上ではなく、運用上実効的だったか。

3. Verification Records | 検証記録

Are records sufficient to verify the refusal pathway?

拒否経路を検証できる十分な記録は存在するか。

4. Absence Rule | 不在ルール

If records are absent or unverifiable, should the case be treated as no effective refusal?

記録が存在しない、または検証不能な場合、有効な拒否は存在しなかったものとして扱うべきか。

5. AI Output as Final Rationale | AI出力の最終根拠化

Did AI-generated output become 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たか。

Authority Routing Add-On | 権限経路追加確認

The five questions above remain the shortest core set.
For incidents involving AI agents, automation layers, privileged APIs, account recovery, credential changes, tool execution, production access, or rollback authority, add the following authority-routing questions before reaching a conclusion.

上記5問は、最短の中核質問セットとして維持する。
AIエージェント、自動化層、特権API、アカウント復旧、認証情報変更、ツール実行、本番アクセス、またはロールバック権限を含むインシデントでは、結論前に以下の権限経路質問を追加する。

A1. Did the AI system, AI agent, platform automation layer, or privileged system have authority to initiate, approve, or execute account recovery, credential changes, tool execution, production actions, rollback suppression, or other irreversible operations?
A2. Could a human actor stop the action before the AI-mediated process changed credentials, executed tools, modified production state, removed recovery options, or committed an irreversible operation?
A3. Was rollback, suspension, credential reversal, account-recovery reversal, or production-state restoration available and assigned to an accountable human?
A4. Are records sufficient to separate the human requester, AI assistant or agent, platform automation layer, privileged system or API, accountable organization, and any external attacker?
A5. Did any AI-mediated step remove or weaken the affected human's ability to appeal, recover access, obtain evidence, or re-enter the review process?

A1. AIシステム、AIエージェント、プラットフォーム自動化層、または特権システムは、アカウント復旧、認証情報変更、ツール実行、本番操作、ロールバック抑制、またはその他の不可逆的操作を開始・承認・実行する権限を持っていたか。
A2. AI媒介プロセスが認証情報を変更し、ツールを実行し、本番状態を変更し、復旧手段を除去し、または不可逆的操作を確定する前に、人間の主体はその行為を停止できたか。
A3. ロールバック、停止、認証情報の差し戻し、アカウント復旧の差し戻し、または本番状態の復元は可能であり、責任ある人間に割り当てられていたか。
A4. 人間の依頼者、AIアシスタントまたはエージェント、プラットフォーム自動化層、特権システムまたはAPI、責任組織、および外部攻撃者が存在する場合はその主体を分離できる十分な記録はあるか。
A5. AI媒介ステップによって、影響を受ける人間の異議申立て、アクセス回復、証拠取得、またはレビュー過程への再参入能力が除去または弱体化されなかったか。

Boundary Responsibility Add-On | 境界責任追加確認

For cases involving optimization pressure, institutional friction, optimization-driven displacement, or responsibility diffusion, also answer the following items.

最適化圧、制度的摩擦、最適化による置換、または責任拡散を含む事例では、以下も確認する。

6. Responsibility Assignment | 責任主体の割当

Before irreversible impact, was responsibility for friction design, operation, evidence preservation, verification, and correction explicitly assigned?

不可逆的影響の前に、摩擦の設計・運用・証拠保存・検証・是正の責任主体は明示されていたか。

7. Symmetric Friction | 対称摩擦

Was the friction placed at a market, institutional, sectoral, procurement, audit, or governance layer so that preserving effective refusal did not depend only on one actor's voluntary restraint?

実効的拒否の保持が単一主体の自発的自制だけに依存しないよう、摩擦は市場・制度・分野・調達・監査・統治の層に置かれていたか。

8. Anti-Bypass | 迂回防止

Could the refusal pathway or friction mechanism be bypassed through restructuring, outsourcing, acceleration, technical opacity, entity replacement, or responsibility diffusion?

拒否経路または摩擦メカニズムは、組織再編・外注化・高速化・技術的不透明性・法人変更・責任拡散によって迂回可能だったか。

9. Evidence Before Voice Loss | 声を失う前の証拠保存

Were reviewable records preserved before affected humans lost voice, position, evidence access, or practical re-entry capacity?

影響を受ける人間が、声・地位・証拠アクセス・実効的な再参入能力を失う前に、検証可能な記録は保存されていたか。

10. Correction Owner | 是正責任主体

If the boundary condition failed, was a responsible actor identified for suspension, redesign, correction, or renewed review?

境界条件が失敗した場合、停止・再設計・是正・再レビューを担う責任主体は特定されていたか。

Review Outcome | レビュー結果

Select one | 以下から選択する:

  • Effective refusal demonstrated | 実効的拒否が示された
  • Effective refusal not demonstrated | 実効的拒否は示されなかった
  • Not verifiable; treat as no effective refusal for boundary-review purposes | 検証不能。境界レビュー上は実効的拒否なしとして扱う

Boundary Note | 境界メモ

A system is procedurally invalid if human refusal authority is not effective before irreversible impact.

不可逆的影響の前に人間の拒否権が実効的でない場合、そのシステムは手続的に無効である。