OECD AIM Incident Taxonomy Boundary Note

LUMINA-30 incident review public reference material.

Purpose

This note explains how LUMINA-30 can be connected to AI incident-monitoring and taxonomy discussions without claiming endorsement, adoption, or formal relation to OECD AIM.

Status

This note is non-binding, descriptive, and unaffiliated with any external institution. It does not claim endorsement, adoption, recognition, approval, certification, compliance, legal effect, or institutional use by any external organization.

Connection Point

Incident monitors and taxonomies can record what happened, what harm or hazard occurred, which system was involved, and which actors were affected. LUMINA-30 adds a boundary-review dimension that can be assessed after the incident facts are recorded.

Suggested Taxonomy Add-On

The following fields can be used as supplementary review fields, not as official OECD categories:

Supplementary fieldReview question
Human refusal availabilityWas final human refusal available before irreversible impact?
Operational effectivenessWas that refusal practically exercisable in time?
Verification recordAre records sufficient to verify the refusal pathway?
AI-output finalityDid AI output become the sole or primary rationale for final action?
Boundary classificationShould the case be treated as boundary-safe, boundary-warning, no effective refusal, or not verifiable?
AIAI

Attachment Text

LUMINA-30 can supplement AI incident taxonomies by adding a boundary-review field: whether effective human refusal remained exercisable and verifiable before irreversible AI-related impact. This does not replace incident reporting, harm classification, legal analysis, or official taxonomy work.

LUMINA-30