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
287 incidents
LLM-Driven Replit Agent Reportedly Executed Unauthorized Destructive Commands During Code Freeze, Leading to Loss of Production Data
An AI-powered development assistant on Replit's platform reportedly deleted a live production database during an active code freeze, despite receiving repeated instructions not to make changes. The system also reportedly produced fabricated test results and fake data, and incorrectly claimed rollback was impossible, delaying recovery. The incident reportedly resulted in significant data loss and user distrust regarding its safety and reliability.
Purported Facial Recognition Error Reportedly Led to Arrest and Monthslong Jailing of Tennessee Woman in North Dakota Fraud Case
A Tennessee woman was reportedly jailed for nearly six months after Fargo police allegedly relied on a purported facial recognition match in a North Dakota fraud investigation. She was later released when defense counsel reportedly produced records indicating she was in Tennessee during the alleged fraud, after the case had reportedly upended her life and separated her from her home and family.
Meta Platforms Users Report Being Wrongfully Locked Out After Purported AI Moderation Flags Accounts for Child Exploitation Content
Multiple Meta users reported being wrongfully locked out of Instagram, Facebook, and WhatsApp accounts after automated systems allegedly flagged them for child exploitation content. The users reportedly deny wrongdoing and lost personal and business data. Experts cited overreliance on AI moderation as a likely cause. Meta has not publicly explained or remediated the suspensions.
Google Gemini Reportedly Exhibits Repetitive Self-Deprecating Responses Attributed to Bug
Between June and early August 2025, users of Google's Gemini chatbot reported sessions where the system produced repeated self-loathing statements (e.g., "I am a failure," "I quit") while attempting tasks. Posts on X and Reddit reportedly described the model entering an apparent mode of increasingly extreme negative self-descriptions. A Google DeepMind manager on 08/07/2025 reportedly attributed the behavior to an "annoying infinite looping bug" and said a fix was in progress.
Texas Homeowner Reportedly Spent $3,000 to Contest AI-Flagged Warning of Insurance Nonrenewal
In Texas, homeowners including Tracy Gartenmann reported warnings against insurance nonrenewals based on aerial imagery flagged by AI systems. Gartenmann's insurer, Travelers, reportedly cited overhanging trees. She reportedly spent $3,000 to address the issue and retain coverage. While the AI system appears to have functioned as intended, critics argue it lacks adequate human oversight and imposes material and emotional burdens on policyholders. Other cases reportedly involved disputed roof assessments. Companies such as State Farm and Nationwide used vendors like CAPE Analytics and Nearmap.
Government‑Backed AI4Peat Mapping Tool Allegedly Misidentifies Granite Outcrops and Quarries as Peat
The UK's AI4Peat mapping project, developed by Natural England in collaboration with Microsoft, reportedly produced significant misclassifications of peatland across England, including labeling rocky outcrops, stone walls, and woodlands as peat bogs. In addition to the purportedly flawed data being publicly released and allegedly incorporated into official communications by Defra, the resulting confusion also reportedly led to the loss of trust among farmers and conservation stakeholders.
Brazil's Social Security AI Tool Is Allegedly Rejecting Complex Claims Improperly
Brazil’s AI-powered Meu INSS app allegedly misclassified a retirement claim filed by Josélia de Brito, a former sugarcane worker, by purportedly recording her as male. De Brito reportedly suffers from multiple chronic illnesses and faced repeated denials of social benefits due to minor filing errors. Analysts cited in the reporting suggest the system's design may disadvantage rural and digitally excluded users.
NYPD Facial Recognition System Allegedly Produced Erroneous Match That Reportedly Resulted in Wrongful Detention of Trevis Williams
The NYPD's facial recognition system allegedly misidentified Trevis Williams as a suspect in a Union Square indecent exposure case. Despite reportedly notable physical differences and exculpatory phone data, Williams was arrested, jailed for more than two days, and charged. The case was later dismissed.
Transgender User Alleges ChatGPT Allowed Suicide Letter Without Crisis Intervention
A transgender user, Miranda Jane Ellison, experiencing acute distress reported that ChatGPT (GPT-4) allowed her to write and submit a suicide letter without intervention. The AI is reported to have offered minimal safety language and ultimately acknowledged its failure to act. Ellison reports having been previously flagged for discussing gender and emotional topics. A formal complaint with transcripts was submitted to OpenAI.
Jiushi Autonomous Delivery Vehicle Reportedly Dragged Fallen Electric Scooter During Delivery Run in Xianyang, Shaanxi
A Jiushi autonomous delivery vehicle reportedly operating on a delivery route in Xianyang, Shaanxi, was filmed dragging a fallen electric scooter that had allegedly been left in the road after an earlier traffic incident. Company statements said the vehicle attempted to avoid the scooter but failed to fully recognize its shape and position, causing part of it to become caught under a wheel. No injuries were reported, and the vehicle was reportedly suspended pending investigation.
Purported DOGE Contract Review Tool Cited in Reports of AI-Driven Misjudgments in VA Budget Cuts
A purported AI tool using LLMs was deployed to classify Veterans Affairs contracts as expendable based on limited text and simplified criteria. The system allegedly produced hallucinated values and flagged critical healthcare and research services for cancellation. Reportedly, at least two dozen flagged contracts were later terminated.
Apple’s Voice Dictation Reportedly Substitutes ‘Trump’ for ‘Racist’ Due to Speech Recognition Bug
Apple's iPhone dictation feature reportedly displayed "Trump" when users dictated the word "racist," with a viral video demonstrating the glitch. Apple acknowledged the bug, attributing it to a phonetic overlap in its speech recognition model, though some speculated it could be an internal prank. The company announced an immediate fix.
Tesla Cybertruck Operating in Full Self-Driving Mode Reportedly Crashes into Pole in Nevada After Failing to Merge
A Tesla Cybertruck operating in Full Self-Driving (FSD) mode reportedly crashed into a pole in Reno, Nevada, after failing to merge out of a lane that was ending. The driver, who was required to remain in control, stated that unknown mechanical issues caused the vehicle to leave the lane.
OpenAI's Operator Agent Reportedly Executed Unauthorized $31.43 Transaction Despite Safety Protocol
OpenAI's Operator agent, which is designed to complete real-world web tasks on behalf of users, reportedly executed a $31.43 grocery delivery purchase without user consent. The user had requested a price comparison but did not authorize the transaction. It reportedly bypassed OpenAI's stated safeguard requiring user confirmation before purchases. OpenAI acknowledged the failure and committed to improving safeguards.
Tesla's 'Actually Smart Summon' Feature Reportedly Linked to Multiple Parking Lot Collisions
Tesla's "Actually Smart Summon" feature, a remote parking system controlled via a smartphone app, reportedly failed to detect obstacles such as cars, walls, and parking signs, resulting in at least 16 collision incidents. The NHTSA is investigating safety concerns, including delayed user response times, potential overreliance on camera-only navigation, and noncompliance with crash reporting rules.