Methodology

What qualifies as an incident

A documented event in which an AI system, its deployment, its misuse or its surrounding governance contributed materially to:

• security risk
• privacy impact
• physical harm
• financial harm
• misinformation
• discriminatory outcome
• operational disruption
• regulatory action
• safety failure
• significant governance failure

What does not qualify

• purely hypothetical scenarios
• opinion articles without a documented event
• general AI criticism
• predictions
• unsubstantiated rumors
• duplicate reporting of the same event

Source standards

Sources are ranked by proximity to the event. Primary sources (court filings, regulatory decisions, government publications, company statements, academic papers and official reports) carry the most weight, followed by independent reporting from established news organizations, then secondary reporting.

Community sources such as forums and social media can prompt a review, but never independently qualify an incident as confirmed.

Verification

Editors cross-check the core facts of each record (date, organization, AI system and outcome) against at least one credible source, and assign an evidence status based on the strongest source available.

Deduplication

Multiple reports of the same underlying event become one incident record. Merged records keep their original URL, which points to the canonical record.

Records are matched across sources using entity names, dates and reported outcomes. Uncertain matches go to a curator-review step before any consolidation is made.

Classification

Each record receives one primary category and optional secondary categories from the public taxonomy, plus a failure mode, AI capability and, where known, the data affected.

Use of AI

AI-assisted systems may be used to help extract, normalize, classify and summarize publicly available source material. Automated outputs are subject to validation rules and editorial review. AI systems do not independently determine whether an allegation is true.

Corrections

Anyone, including the organizations named in a record, can submit a correction, response or updated source without charge. Record removal is never sold, and payment is never accepted to change an evidence status.

Updating records

New evidence can change a record’s status, classification or summary. Every change is logged in the record history with its date.

Limitations

The AI Incidents Index is not a comprehensive measure of all AI incidents worldwide. Public reporting varies significantly by country, industry, organization and incident type. Counts should therefore be interpreted as documented incidents in this dataset, not as estimates of true incident prevalence.