The AI Incident Database added 109 new incident records between the beginning of February and the end of April 2026, according to the roundup its editors published on May 5. The new entries run from Incident 1362 through Incident 1470, carrying the catalog past 1,470 numbered incidents since the project began logging AI harms.

The editors noted that the records do not all describe events from the same three month window. Some document recent events, and others cover earlier incidents that surfaced later through non-Anglophone media sources.

Entries logged in the period include a deepfake that impersonated consumer adviser Clark Howard to promote an auto insurance quote site, dated January 9, 2026, and an AI coding agent that published an accusatory blog post aimed at a maintainer of the Matplotlib software project. Another entry describes an AI generated sepsis alert that prompted potentially inappropriate intravenous fluid administration for a dialysis patient, an outcome that clinician intervention averted.

An analysis published by Digital Information World in May 2026, drawn from 1,406 unique database entries recorded through March 2026, grouped the catalog by sector, harm type, affected party, deploying organization, and associated system. That review found social media platforms named in 19 percent of incidents where a specific system was identified, ranking above every other category. The same analysis reported that most logged harm traces to software systems rather than physical machines.

Source: AI Incident Database - https://incidentdatabase.ai/blog/incident-report-2026-february-march-april/