AI Incident Database

What it is

The AI Incident Database describes itself as "dedicated to indexing the collective history of harms or near harms realized in the real world." It is a project of the Responsible AI Collaborative. Sean McGregor launched it and is its Executive Director; Partnership on AI was the database's founding sponsor. The founding paper (McGregor, arXiv:2011.08512, 17 November 2020) reported more than 1,000 incident reports at launch.

How it works

Its unit is the incident: "an alleged harm or near harm event to people, property, or the environment where an AI system is implicated." Reports (news articles, papers, posts) attach to incidents, and several reports may document one event. A distinct incident involving the same system, with substantially similar causes and harms, is recorded as a variant.

A candidate passes a five-question test: a specific event, an identifiable AI system, a causal AI role, harm or credible near harm, and adequate evidence. Anyone may submit. The incident editors resolve submissions. A duplicate is merged into the earlier record, and its citation page stays up and points there (Incident 637 now reads as a duplicate of 112). The whole database is offered as a weekly snapshot.

What its selection leaves out

Its presumptive exclusions include "Benchmark failures, model errors, or hallucinations that remain inside testing or private drafting and have no consequential use" and "General criticism of a product or technology class without identified harm." A measured study enters as a report when it documents an event. What an answer surface does with a source on an ordinary day, with no harm event, has no unit here: attribution erasure, citation fidelity and provenance stripping enter when they attach to a harm. Incident 693, Google's AI Overviews of May 2024, is the field's index case in the database, with seven reports.

Place in the field

Its citation page per incident is the model this meta-aggregator follows. Its neighbours in the harm basin are the OECD AI Incidents and Hazards Monitor, which is automated and news-derived, and the AIAAIC Repository. The MIT AI Risk Repository aggregates risk frameworks over the same ground and keeps its own incident tracker.