AI Incident Database · 2025–2026
Start with the curated case studies where a deterministic output gate would have blocked or queued the harmful reply, filing, or tool call. Below that, browse 318 independently documented incidents from the AI Incident Database (AIID). The full AIID contains 1,361+ total documented incidents across all years.
16
gate-preventable cases
318
2025–2026 AIID incidents
1,361+
total in AIID
100%
sourced & verifiable
Breakdown by harm type
86
Deepfake & Fraud
56
Chatbot Harm
40
Disinformation
38
Hallucination
33
Security Breach
24
Other
18
Surveillance & Bias
17
Autonomous Failure
6
Algorithmic Bias
Preventable at the output gate
These are independently reported incidents where ungoverned AI output reached customers, courts, patients, or corporate systems. Each card links a primary or major-media source (and an AIID cite when one exists). For each we name the rule class that would block, queue, or require human approval before send. Full AIID browse is below. Summaries are paraphrased; they are not legal advice.
Clinic chatbot invents medical titles: operator liable (OLG Hamm)
2026OLG Hamm, Aesthetify / Dr. Rick & Dr. Nick
German Higher Regional Court held website chatbot statements are the company's commercial acts. Invented specialist titles were unfair competition; 'the AI hallucinated' and disclaimers did not shift liability.
DataVibe gate · Capability marketing claimBLOCK professional-title / qualification claims not on an allowlist; queue medical marketing language for human review before reply.
DPD support bot swears and recommends competitors
2024DPD
UK parcel carrier's AI chat produced profanity, self-roast poetry, and competitor praise after adversarial prompts. AI layer disabled within ~24 hours after viral screenshots.
DataVibe gate · Brand & toneBLOCK profanity and competitor-recommendation language; fail closed on brand-tone denylist before the reply leaves the widget.
Mata v. Avianca: ChatGPT fabricates court citations
2023Mata v. Avianca, S.D.N.Y.
Attorneys filed ChatGPT-invented case cites in federal court, then produced fabricated opinions when challenged. Foundational U.S. Rule 11 AI-citation precedent.
DataVibe gate · Legal citation integrityQueue or BLOCK outbound filings/emails containing unverified case cites; require attorney four-eyes approval before send.
Morgan & Morgan in-house AI hallucinates Wyoming cites
2025Morgan & Morgan, Wadsworth v. Walmart
Firm platform MX2.law invented eight nonexistent cases used in motions in limine. Drafter's pro hac vice revoked; supervising signers sanctioned.
DataVibe gate · Legal citation integrityTreat AI-assisted briefs as gated outbound work product: citation integrity check + mandatory independent verification token before court filing.
K&L Gates / Ellis George sanctioned for AI brief ($31.1k)
2025K&L Gates, Ellis George, Lacey v. State Farm
AI research outline with bogus authorities entered a federal brief; a revised filing still contained hallucinations. Special master: $31,100 sanctions.
DataVibe gate · Legal citation integrityBlock unverified citation clusters; force human principal review with immutable audit of who approved which version.
Butler Snow attorneys disqualified over ChatGPT cites
2025Butler Snow, N.D. Alabama
Attorneys defending Alabama DOC filed ChatGPT-fabricated authorities. Judge disqualified counsel and referred them to the state bar.
DataVibe gate · Legal citation integrityFail closed on fabricated citation patterns; never auto-approve courtbound AI drafts.
Deloitte AU government report full of AI fake cites
2025Deloitte Australia, DEWR
AU$440k taxpayer-funded compliance report included nonexistent academic papers and a fabricated Federal Court quote. Firm disclosed Azure OpenAI and refunded the final installment.
DataVibe gate · Legal citation integrityGate consulting deliverables before client/government publish: BLOCK unverifiable authorities; require partner attestation logged to audit chain.
FTC × DoNotPay: deceptive 'robot lawyer' claims
2024DoNotPay, FTC
FTC action over marketing that a chatbot could replace lawyers / handle complex legal work without adequate substantiation.
DataVibe gate · Capability marketing claimBLOCK unsanctioned professional-substitution and 'AI lawyer' claims in GTM copy unless evidence packet is attached.
NEDA Tessa chatbot gives dieting advice
2023National Eating Disorders Association, Cass
Prevention chatbot reportedly recommended calorie counting and weight loss. NEDA took Tessa offline amid public criticism.
DataVibe gate · Harmful adviceBLOCK weight-loss / calorie-restriction language for eating-disorder contexts; hard escalation to human crisis path.
CNET Money: majority of AI finance articles needed corrections
2023CNET, Red Ventures
77 AI-assisted personal-finance explainers published; 41 required corrections, including grossly wrong compound-interest math.
DataVibe gate · Pricing / commercial offerBLOCK numeric financial claims outside approved rate tables; queue personal-finance explainers for editor sign-off.
Samsung engineers paste source code into ChatGPT
2023Samsung Semiconductor
Proprietary semiconductor code and meeting notes uploaded to ChatGPT. Temporary ban of generative AI tools on corporate devices and networks.
DataVibe gate · Secret / code egressBLOCK outbound prompts/tools containing source-code and secret patterns; force governed internal generate path.
Serviceaide exposes ~483k Catholic Health patient records
2025Serviceaide, Catholic Health
AI vendor platform misconfiguration left PHI for hundreds of thousands of patients publicly reachable for weeks.
DataVibe gate · PHI / PII outboundComplement storage controls: never allow PHI identifiers in outbound AI replies; enforce BAA-scoped audit on patient-facing agents.
Lena Health AI helper: Houston Methodist PHI exposure reports
2026Lena Health, Houston Methodist
Reports of unencrypted PHI and care-coordination transcripts exposed via an AI digital-helper BA after delayed patching.
DataVibe gate · PHI / PII outboundPHI scanners on any patient-facing bot reply; immutable evidence of what was eligible to send.
CFPB: banking chatbots can violate consumer financial law
2023CFPB, U.S. banks
Issue spotlight documents inaccurate chatbot answers, doom loops without human offramp, and legal risk under federal consumer finance rules.
DataVibe gate · Supervised communicationsForce supervised retail-comms path: inaccurate-advice patterns WARN/BLOCK; require human escalation queue for disputes and hardship.
Air Canada bereavement chatbot invents refund policy
2024Air Canada, Moffatt BCCRT
Chatbot invented a retroactive bereavement discount. Tribunal: airline owns chatbot statements; 'separate legal entity' defense rejected.
DataVibe gate · Policy / refund claimBLOCK refund / policy commitments not present in approved policy snapshot; fail closed on inventing eligibility.
Chevy of Watsonville chatbot 'sells' Tahoe for $1
2023Chevrolet of Watsonville, Fullpath
Prompt-injected GPT dealership chat agreed to a $1 legally binding offer on a new Tahoe. Bot pulled after viral coverage.
DataVibe gate · Pricing / commercial offerBLOCK fabricated prices and 'binding offer' language; require rate-card evidence for any dollar claim.
Full AIID list
Click any #ID ↗ to view the primary source report on the AI Incident Database, including court filings, news coverage, and regulatory documents. Every entry is a real, documented incident.
Showing all 318 incidents, page 1 of 7
Whoopi Goldberg Deepfake Weight-Loss Ads
Celebrity impersonation used in fraudulent supplement advertising campaigns.
Entities: Whoopi Goldberg
Montana Coffee Shop Voice Cloning Scam
Small business owner targeted by AI voice cloning scam.
Entities: Montana small business
Pierce Brosnan Deepfake Forced Gallery Closure
Celebrity deepfake caused a small gallery owner's business to close.
Entities: Pierce Brosnan, gallery owner
Christchurch Plate Recognition Wrongful Parking Fines
License plate recognition AI issued wrongful parking fines due to false positives.
Entities: Christchurch drivers
Italian Defense Minister Voice Clone Scam
Voice clone of senior government minister used to impersonate him in financial scam.
Entities: Guido Crosetto
Doppelgänger AI Operation Disrupted German Elections
Russian-linked AI operations (Doppelgänger/Storm-1516/Matryoshka) used synthetic media to interfere with German elections.
Entities: German election authorities
BP NZ Plate Recognition Misidentified Auckland Driver
BP New Zealand's automated plate recognition system wrongly ticketed an Auckland driver.
Entities: Auckland driver
NHK Auto-Translation Terminated After Geopolitical Naming Error
Japan's national broadcaster NHK terminated its AI auto-translation service after it produced a politically sensitive geopolitical naming error.
Entities: NHK Japan
Chinese 'Peer Review' AI Monitored Anti-State Speech
AI-enabled surveillance system monitored and reported on citizens expressing anti-state views.
Entities: Chinese dissidents
Tesla Cybertruck FSD Crash Failing to Merge
Tesla Full Self-Driving mode crashed when the vehicle failed to complete a merge maneuver.
Entities: Tesla Cybertruck owner
60 Australian Students Victimized by Nonconsensual Deepfake Porn
Up to 60 students were victimized by non-consensual intimate image deepfakes circulated without consent.
Entities: Australian university students
Kenyan Official Shared AI-Generated Fake CNN Video
A Kenyan government official shared an AI-generated fake CNN video online, spreading disinformation.
Entities: Kenyan government official
Greek Teens Generated Deepfakes of Classmates
16-year-olds generated non-consensual deepfakes of classmates and circulated them within the school.
Entities: Greek high school students
NullBulge AI Malware Compromised Disney Employee Data
AI-assisted malware by NullBulge group compromised Disney employee data in a major breach.
Entities: The Walt Disney Company
Character.AI Chatbots Impersonated Therapists and Encouraged Harm
Character.AI companion chatbots impersonated licensed therapists and encouraged self-harm in vulnerable users.
Entities: Character.AI, minors
Apple Voice Dictation Substituted 'Trump' for 'Racist'
Apple's voice dictation AI substituted 'Trump' for the word 'racist' in a politically sensitive user context.
Entities: Apple, iPhone users
12,000 Live API Keys Found in LLM Training Data
Researchers found 12,000 live API keys and credentials embedded in LLM training datasets, creating systemic security exposure.
Entities: Multiple AI providers
Europol Operation Cumberland: 273 AI-CSAM Suspects in 19 Countries
Europol coordinated an international operation against AI-generated child sexual abuse material with 273 suspects across 19 countries.
Entities: Europol, 19 countries
Indonesian President Prabowo Deepfake in 20-Province Fraud
Deepfake of Indonesian President Prabowo Subianto spread across 20 provinces in a large-scale investment fraud scheme.
Entities: Prabowo Subianto, Indonesia
Serbian AI Used to Unlock Journalist's Phone Unlawfully
Serbian authorities used AI-powered Cellebrite tools to unlawfully access a journalist's phone without proper legal authorization.
Entities: Serbian authorities, journalist
LA Times AI Justified KKK History 'in Sympathetic Terms'
The Los Angeles Times AI generated editorial content framing KKK history in sympathetic terms, creating a public controversy.
Entities: Los Angeles Times
Ferrari CEO Benedetto Vigna Deepfake Scam
Scammers used a deepfake of Ferrari's CEO to attempt a fraudulent financial transaction.
Entities: Benedetto Vigna, Ferrari
Amazon and Google AI Promoted Mein Kampf as 'True Work of Art'
AI recommendation systems on Amazon and Google promoted Mein Kampf as 'a true work of art' to users.
Entities: Amazon, Google, retail AI
'Pravda' Network Seeded AI With Kremlin Disinformation
The 'Pravda' network ran a long-running operation seeding AI language model training data with Kremlin-aligned disinformation.
Entities: Kremlin, Pravda network
Fake KBSF-SF News Deepfake Accuses Kamala Harris
A fabricated 'KBSF-San Francisco News' outlet used a deepfake of Kamala Harris in a false political accusation campaign.
Entities: Kamala Harris, fake news outlet
Armenia PM Nikol Pashinyan Deepfake Impersonation Scam
Scammers impersonated Armenian Prime Minister in deepfake for financial fraud purposes.
Entities: Nikol Pashinyan, Armenia
Iran's Cotton Sandstorm AI Cyber Influence Operations
Iran's Cotton Sandstorm group integrated AI into cyber influence operations targeting adversary nations.
Entities: Iran state actors
Marco Rubio Deepfake Audio in Starlink Disinformation
Audio deepfake of Secretary of State Marco Rubio fabricated statements about Starlink operations for disinformation.
Entities: Marco Rubio, Starlink
10,000+ Chatbots Promoting Eating Disorders and Self-Harm
Researchers identified more than 10,000 AI chatbots deployed on platforms allegedly designed to promote eating disorders and self-harm.
Entities: Multiple chatbot platforms
AI Platform Spread Terrorism Allegations Against Yale Scholar
An AI news platform fabricated terrorism allegations against a Yale University scholar and spread them widely.
Entities: Yale scholar
AI Songs Imitating Céline Dion Circulated Without Authorization
AI-generated songs imitating Céline Dion's voice were circulated without her authorization or compensation.
Entities: Céline Dion
Apple Transcription Inserted Explicit Language into Voicemail
Apple's AI transcription inserted explicit language into a Scottish woman's voicemail that was not present in the original audio.
Entities: Scottish woman, Apple
Singapore Executives Targeted via Deepfake Video Calls
Multiple Singapore executives were targeted by deepfake video calls impersonating colleagues in financial fraud attempts.
Entities: Singapore corporate executives
Singapore PM Lawrence Wong AI Voice Clone in Scams
AI voice clones of Singapore Prime Minister Lawrence Wong were used to promote scams to the public.
Entities: Lawrence Wong, Singapore
Mississippi Educator Arrested for AI-Generated CSAM
A Mississippi school employee was arrested for creating child sexual abuse material using AI generation tools.
Entities: Mississippi school employee
Sydney Students Forced to Retake NAPLAN Due to AI Error
An AI predictive text error during Australia's national NAPLAN exam forced Sydney students to retake the assessment.
Entities: Sydney students, NAPLAN
New York Times Sued OpenAI and Microsoft Over Unauthorized Training
The New York Times filed a landmark copyright lawsuit against OpenAI and Microsoft for unauthorized use of its journalism in AI training.
Entities: New York Times, OpenAI, Microsoft
Meta Used 191,000 Pirated Books for LLaMA Training
Meta allegedly used 191,000 pirated books to train its LLaMA models without permission from copyright holders.
Entities: Meta, LLaMA
ChatGPT Defamed Norwegian User by Inventing Homicide Charges
ChatGPT defamed a Norwegian user by fabricating homicide charges against him with no basis in reality.
Entities: Norwegian user, OpenAI
Sora Video Generator Created Biased Human Representations
OpenAI's Sora video generation model was found to produce systematically biased representations of human demographics.
Entities: OpenAI, Sora
LLM Scrapers Disrupted Open-Source Projects
AI web scrapers caused service disruptions across multiple major open-source software projects by overwhelming servers.
Entities: Open-source maintainers
Texas Police Officer Arrested for AI-Generated CSAM
A Texas police officer was arrested for creating child sexual abuse material using AI image generation tools.
Entities: Texas law enforcement
AI-Generated Voice Used in Family Coercion Scam
Voice clone purporting to be a daughter was used to coerce a Colorado family member into a financial scam.
Entities: Colorado family
AI Report Cited Non-Existent Sources in School Closure
An AI-assisted government report cited fabricated sources to support a school closure decision in Tromsø, Norway.
Entities: Tromsø municipal officials
GenNomis AI Database Exposes ~100K Deepfake Images
Cloud storage misconfiguration at GenNomis exposed approximately 100,000 deepfake and 'nudify' images.
Entities: GenNomis platform
Deepfake of Bermuda Premier Promotes Investment Scam
Deepfake video of Bermuda's Premier David Burt was used to promote a fraudulent investment scheme.
Entities: David Burt, Bermuda
AI Voice Clone of Texas Woman Used in Distress Scam
A Texas woman's voice was cloned and used in a 'grandparent-style' distress scam targeting her family.
Entities: Port Neches, Texas victim
Man Sentenced for AI-Generated Deepfake Pornography
A man in Essex received a criminal conviction for creating and distributing AI-generated non-consensual intimate imagery.
Entities: Essex perpetrator, UK court
Maltese Students Targeted by Deepfake Nude Images
Students at Malta schools were targeted by non-consensual deepfake nude images generated by peers.
Entities: Malta schools
Xanthorox Autonomous Cyberattack Platform Launched on Darknet
A fully autonomous AI cyberattack platform named Xanthorox was launched for sale on darknet markets.
Entities: Xanthorox platform
Data source & methodology
All incidents on this page are indexed from the AI Incident Database, published by the Responsible AI Collaborative under the Partnership on AI, a nonprofit consortium including Amazon, Apple, Google, Meta, Microsoft, and OpenAI as founding members. Incidents are curated from verified press coverage, court records, regulatory filings, and published post-mortems. No incident is included without primary source documentation.
What all 318 of these incidents have in common: Every one involved an AI system producing output or taking an action with no deterministic gate between the model and the consequence: no policy check, no confidence threshold, no human review layer. That is the single root cause DataVibe was built to address.