AI Incidents
← Back to incident register

Transparency

Methodology over headlines.

AI Incidents separates observable facts, editorial assessment, and unverified claims. Every published record links to its evidence.

01

Concrete event

A record needs a clear time reference, an affected organization or technology, and a describable impact.

02

Credible sources

At least one primary or authoritative source is required. For disputed or severe claims, we aim for two independent sources.

03

Clear separation

Observation, interpretation, and unverified claims are stored separately and shown as distinct information.

04

Versioned review

Confidence, verification status, first observation, disclosure, and last review remain traceable as separate fields.

Editorial workflow

From signal to published incident

  1. DiscoveredA possible incident is recorded as a candidate, not published.
  2. TriagedRelevance, event scope, and possible duplicates are reviewed.
  3. EvidencedSources support specific claims while uncertainty remains visible.
  4. ApprovedPublication requires a documented editorial review.
  5. UpdatedCorrections and new sources transparently update the versioned record.

Required fields

Confidence is not a truth score

The confidence value describes the current strength of a record's evidence. It replaces neither sources nor editorial review and must not be read as the probability that a headline is true.

verificationStatus
Working state from demo or needs evidence through verified, corrected, or withdrawn.
sourceType
Primary source, authority, research, secondary source, or clearly marked demo source.
lastReviewedAt
Time of the latest editorial review, not the event date.