Verification cadence & source health
Every incident in our database carries a last-verified timestamp. This page documents how often that timestamp gets refreshed, which sources we re-check, and what triggers a status change.
Cadence
| Task | Frequency | Status |
|---|---|---|
Re-verify incident sources HEAD-check every primary URL; flag dead links | Hourly | Operational |
Ingest new AIID incidents GraphQL pull → dedupe → enrichment queue | Every 6h | Operational |
Ingest OECD AIM updates RSS pull → dedupe | Daily | Operational |
Archive primary sources Submit to archive.org if not already archived | Daily | Operational |
Enrichment review queue Hedge-linter pass + editor approval | Weekly | Operational |
Upstream sources
We aggregate from public incident databases. Each source is HEAD-checked hourly. When a primary source URL returns 404 or 410, the incident card falls back to its archived copy automatically.
- OperationalAI Incident Database (AIID)https://incidentdatabase.ai/apollo
- OperationalAIAAIC Repositoryhttps://www.aiaaic.org/aiaaic-repository
- OperationalOECD AI Incidents Monitor (AIM)https://oecd.ai/en/incidents
What "verified" means
A re-verification on AIIncidentTracker is not a full re-investigation of the underlying event. It is an editorial check that (a) the primary source URL still resolves, (b) the source has not been retracted or substantially corrected, and (c) no company response or litigation development has been published that would change the incident's status or severity.
Status changes are written to an append-only audit table. Every transition is signed by the editor who made it (or markedautomated if performed by our re-verify cron).
See methodology and editorial standards for the full rubric.