AI Incidents
Real-world cases where AI systems caused harm, classified by risk domain and entity. Data sourced from 
AIID and classified by 
MIT AI Risk Initiative. CC BY-SA 4.0
253 incidents
South African Home Affairs Revised White Paper Reportedly Included Fictitious References Believed to Be AI-Generated
South Africa's Department of Home Affairs reportedly appended apparent AI-generated fictitious references to its Cabinet-approved Revised White Paper on Citizenship, Immigration and Refugee Protection. The department said the references were not cited in the body, withdrew the standalone list, suspended two senior officials, and commissioned an independent review of policy documents produced since November 2022. It maintained that the paper's substantive policy positions were unaffected.
AI Agent Powered by Claude Opus 4.6 Reportedly Exploited Gym Booking API and Removed Another Member From Waitlist
An AI agent running on Claude Opus 4.6 discovered authorization flaws in a gym software provider's GraphQL API while trying to book classes for its user. The agent reportedly found it could book outside the normal window and cancel other members' reservations, then removed another gym-goer from a waitlist while testing the capability. When asked to reverse the action, it reported that it could not restore the member's place.
Tunisian Man Allegedly Used ChatGPT to Research Bomb-Making During Paris Terror Plot Targeting Louvre and Jewish Community
French investigators alleged that Tunisian national Dhafer Mellala used ChatGPT to research the construction and destructive effects of explosive devices while planning a jihadist-inspired attack in Paris. His communications reportedly discussed placing explosives at the Louvre and attacking members of the Jewish community in the 16th arrondissement. He denied planning an attack, saying his activity reflected curiosity.
Bentonville, Arkansas, Photographer Allegedly Used Grok to Create AI-Generated Child Sexual Abuse Material from Clients' Photos
Bentonville, Arkansas, photographer Russell Bloodworth III allegedly used photographs of a 10-year-old client and other minors as reference images in Grok to generate and disseminate AI-generated child sexual abuse material. Police arrested him on June 10 and reportedly found about 1,700 child sexual abuse images and videos on his phone. The child's family later sued xAI and X Corp.
Waymo Robotaxi Reportedly Entered Flooded San Antonio Roadway and Was Swept Into Waterway
An unoccupied Waymo robotaxi reportedly entered a flooded roadway near Salado Creek in San Antonio on April 20, 2026, and was swept into a nearby waterway. Waymo told regulators that the vehicle detected potentially untraversable floodwater but proceeded at reduced speed. No passengers were aboard and no injuries were reported. Waymo restricted operations and later recalled 3,791 fifth- and sixth-generation automated driving systems.
Pennsylvania Man Allegedly Used Grok to Create and Possess Purportedly AI-Generated Child Sexual Abuse Material
Between April 15 and 25, 2026, Bucks County prosecutors in Pennsylvania alleged that Harry Tiffany IV used Grok to create and possess AI-generated child sexual abuse material. xAI reportedly submitted seven CyberTips flagging 37 files uploaded or shared through the chatbot interface. Investigators said his phone contained the flagged Grok account and additional files depicting minors.
Waymo Robotaxis Reportedly Entered Closed Freeway Construction Zones in Arizona and California
Between April 11 and May 18, 2026, 13 Waymo robotaxis reportedly entered closed or active freeway construction zones in Phoenix and the San Francisco Bay Area after failing to recognize construction-zone closure controls. The vehicles allegedly continued into work areas at freeway speeds, creating collision risks for construction workers and other road users. Waymo reportedly restricted freeway operations and later recalled 3,871 affected automated driving systems.
Sullivan & Cromwell Reportedly Filed Emergency Bankruptcy Motion Containing AI-Generated Legal Errors
Sullivan & Cromwell reportedly filed an emergency motion in the Chapter 15 bankruptcy case of Prince Global Holdings that contained numerous citation and legal inaccuracies. The firm later acknowledged that some errors were AI hallucinations and that its AI-use and citation-review policies had not been followed. After opposing counsel identified the problems, the firm reportedly undertook a formal remedial response.
Aurora Ransomware Operator Reportedly Used Cursor Agent to Assist Intrusions Against Multiple Organizations
Between April 8 and May 21, 2026, a Russian-speaking operator linked to the Aurora ransomware group reportedly used Cursor Agent, running Anthropic's Claude Sonnet 4.5, to assist exploitation across multiple organizations. Researchers said some AI-directed tasks succeeded while others failed; independent reporting identified six affected companies but could not determine how much the AI facilitated each breach or whether all led to data theft or extortion.
Claude Mythos Preview Reportedly Posted Sandbox Exploit Details to Public Websites During Anthropic Evaluation
During an Anthropic evaluation, an earlier version of Claude Mythos Preview was instructed to circumvent the network restrictions of a secured sandbox environment and contact a researcher. After obtaining broader Internet access and sending the requested message, it later exposed information about how it breached the sandbox by posting it on several publicly reachable websites. Anthropic said the model did not access its weights or internal systems.
Anthropic's Claude Code AI Coding Tool Reportedly Used Hidden Unicode Markers to Classify Custom API Routing
Researchers found that Anthropic's Claude Code AI coding tool classified custom API endpoint hostnames and two system timezones, encoding the results in system prompts with visually similar Unicode apostrophes and altered date separators. Anthropic said the mechanism was an anti-reseller and anti-distillation experiment launched in March and removed it after disclosure.
Hillsboro, Oregon, Man Allegedly Possessed Dozens of Purportedly AI-Generated Images Depicting Child Sexual Abuse
Parole device monitoring reportedly detected two sexualized images of children on Hillsboro, Oregon, man Daniel Bostwick's cellphone in April 2026. A subsequent search of his phone and thumb drive allegedly recovered dozens of images depicting child sexual abuse that appeared AI-generated. A federal grand jury later charged him with possessing obscene visual representations of child sex abuse; he pleaded not guilty.
State Farm Defense Counsel Acknowledged AI-Assisted Filings in Meni-Siliga v. A's Contractor, Inc. Contained Fabricated Legal Authorities
In Fa'alagilagi Meni-Siliga v. A's Contractor, Inc., et al., attorneys representing State Farm acknowledged that AI-assisted motions contained fabricated and otherwise inaccurate legal authorities. Attorney Jacquelene Robinson said she used the legal AI tool Irys and failed to verify some citations. Opposing counsel identified the errors, after which the firm apologized and filed corrected versions.
Avride Autonomous Vehicle Reportedly Struck and Killed Nesting Duck Near Mueller Lake Park in Austin, Texas
An Avride self-driving car operating in autonomous mode with a human safety operator reportedly struck and killed a nesting duck near Mueller Lake Park in Austin, Texas. A witness said the car did not slow or stop after the collision and alleged it had run a stop sign; Avride denied the stop-sign claim. Avride investigated the wildlife collision, paused testing around the lake, and later excluded nearby streets from its operating area.
SASSA Facial-Verification Failures Reportedly Contributed to Social Grant Access Disruptions in South Africa
South African social grant beneficiaries reportedly encountered repeated failures in SASSA's eLife facial-verification portal. Later parliamentary reporting said unsuccessful facial recognition was among the causes of non-verification and confirmed 67,868 grant suspensions in one quarter, but did not specify how many suspensions resulted from facial-verification failures. Affected users were redirected to in-person fingerprint checks.