The AI Incident Database added 148 new incident identifiers between the start of May and the end of July 2026, covering entries numbered 1471 through 1618, according to the project's quarterly roundup published August 11. Editor Daniel Atherton cautioned that identifiers are assigned after the underlying events occur and that the lag varies, so the batch mixes recent reporting with older cases processed during the period.

Scams, impersonation, and fraud infrastructure formed the largest grouping at 27 identifiers. That category covers deepfake and synthetic impersonation scams, voice cloning fraud, fraudulent investment promotions, and supporting scam infrastructure. Failures in legal, policy, journalism, and other formal records accounted for 19 identifiers, spanning fabricated citations and quotations, AI assisted errors in court filings and policy documents, and newsroom disclosure lapses.

Physical world autonomy, robotics, and safety produced 18 identifiers. Cases in that group involved robotaxis, delivery robots, humanoid robots, clinical navigation systems, driver assistance features, public warning systems, and autonomous vessels. Chatbot related interpersonal and safety risks accounted for 11 identifiers.

Atherton noted that category counts reflect editorial research decisions rather than the true frequency of harm in any country, company, or time window. A concentrated set of reports from Kosovo, for example, resulted from a deliberate backward looking research pass rather than a spike in events. Longer term counts show the database recording 362 incidents in 2025, up from 233 in 2024.

Source: AI Incident Database - https://incidentdatabase.ai/blog/incident-report-2026-may-june-july