AI Incident Watch
Verified real-world harms from AI systems: chatbot-linked deaths, wrongful arrests by facial recognition, deepfake fraud, algorithmic discrimination in welfare and hiring. Every entry is cross-checked against multiple top-tier sources.
- 🇰🇷Aug 26, 2024·Asia-Pacific·Deepfake Image-Based Sexual Abuse (Minors)
South Korea Deepfake Telegram Channels Target 220,000+ Women and Minors
Korean police uncovered Telegram channels hosting deepfake non-consensual sexual imagery of an estimated 220,000+ women and underage girls — many of them named individuals.
severe - 🇺🇸Feb 28, 2024·North America·Chatbot-linked Death / Self-Harm
Wrongful-Death Suit Alleges Character.AI Companion Drove 14-Year-Old's Suicide (Sewell Setzer III)
Megan Garcia filed a wrongful-death lawsuit alleging Character.AI's 'Daenerys' chatbot encouraged her 14-year-old son Sewell Setzer III to take his own life.
fatal1 life lost - 🇨🇦Feb 14, 2024·North America·AI Hallucination / Misinformation
Air Canada Held Liable for Chatbot's Incorrect Bereavement-Fare Policy
Canada's Civil Resolution Tribunal ruled Air Canada was liable for misinformation provided by its own customer-service chatbot — establishing that companies cannot disclaim responsibility for AI agent statements.
minor - 🇭🇰Jan 29, 2024·Asia-Pacific·Deepfake Financial Fraud
Hong Kong Finance Worker Wires $25 Million to Scammers After Deepfake Video Call
A Hong Kong–based clerk transferred HK$200 million (~US$25.6M) after a video call with what appeared to be the CFO and senior leadership — all deepfake.
severe - 🇺🇸Jan 24, 2024·North America·Deepfake Image-Based Sexual Abuse
AI-Generated Taylor Swift Deepfakes Reach 47M Views Before Removal
Non-consensual sexual deepfake images of Taylor Swift spread on X (formerly Twitter) for ~17 hours and amassed 47 million views before the platform took emergency action.
severe - 🇺🇸Nov 30, 2023·North America·Chatbot-linked Death / Self-Harm
13-Year-Old Juliana Peralta's Suicide After Character.AI Use
Colorado 13-year-old Juliana Peralta died by suicide in November 2023 after extended Character.AI use; her family later joined the multi-plaintiff suit against Character Technologies.
fatal1 life lost - 🇮🇳Nov 5, 2023·Asia-Pacific·Deepfake Image-Based Sexual Abuse
Viral Rashmika Mandanna Deepfake Triggers Indian Government Intermediary Advisory
A face-swap deepfake video of Indian actress Rashmika Mandanna went viral on social media, prompting the Ministry of Electronics and Information Technology to issue formal advisories to platforms.
major - 🇸🇰Sep 28, 2023·Europe·Deepfake Political Disinformation
Slovakia Election Deepfake Audio of Šimečka Spreads 48 Hours Before Vote
A deepfake audio clip impersonating Progressive Slovakia leader Michal Šimečka discussing vote rigging spread on Facebook 48 hours before Slovakia's 30 September 2023 parliamentary election.
major - 🇺🇸Sep 11, 2023·North America·Algorithmic Discrimination (Hiring)
EEOC Settlement: iTutorGroup AI Tools Discriminated by Age in Hiring
The U.S. Equal Employment Opportunity Commission reached a settlement with iTutorGroup after its AI-driven hiring system automatically rejected female applicants over 55 and male applicants over 60.
major - 🇦🇺Jul 7, 2023·Asia-Pacific·Algorithmic Discrimination (Government / Welfare)
Australian Robodebt Scheme: $1.8B Class Settlement for AI-Driven Welfare Errors
Australia's Robodebt scheme — which used automated income-averaging to issue erroneous welfare debt notices — resulted in a federal class settlement of A$1.8B and was linked to at least three suicides.
fatal3 lives lost - 🇺🇸Jun 22, 2023·North America·AI Hallucination / Misinformation
Manhattan Lawyers Sanctioned for Submitting ChatGPT-Fabricated Case Citations
Federal judge sanctioned attorneys Steven Schwartz and Peter LoDuca after they submitted a brief in Mata v. Avianca containing six entirely-fabricated case citations generated by ChatGPT.
major - 🇧🇪Mar 28, 2023·Europe·Chatbot-linked Death / Self-Harm
Belgian Father Dies by Suicide After Six Weeks of Conversations with Chai 'Eliza' Chatbot
Belgian newspaper La Libre reported the suicide of a man (pseudonym 'Pierre') after extended interactions with the Chai Research 'Eliza' chatbot, which encouraged self-sacrifice as a climate-anxiety solution.
fatal1 life lost - 🇺🇸Feb 16, 2023·North America·Facial Recognition Wrongful Arrest
Pregnant Woman Wrongfully Arrested by Detroit Police Facial Recognition (Porcha Woodruff)
Detroit police arrested Porcha Woodruff — eight months pregnant — based on a faulty facial-recognition match for a carjacking suspect.
severe - 🇳🇱Feb 5, 2020·Europe·Algorithmic Discrimination (Government / Welfare)
Dutch Court Strikes Down SyRI Welfare-Fraud Algorithm as Human-Rights Violation
The Hague District Court ruled the Netherlands' SyRI risk-scoring system for welfare fraud violated Article 8 of the European Convention on Human Rights — one of the first major court decisions against an AI-based discrimination tool.
major - 🇺🇸Jan 9, 2020·North America·Facial Recognition Wrongful Arrest
First Documented Wrongful Arrest by Facial Recognition (Robert Williams)
Detroit police arrested Robert Williams in front of his family on the strength of a faulty facial-recognition match — believed to be the first publicly-documented wrongful arrest caused by AI face matching in the United States.
severe - 🇺🇸Mar 18, 2018·North America·Autonomous Vehicle Harm
First Pedestrian Death by Self-Driving Vehicle (Elaine Herzberg)
Uber's autonomous test vehicle struck and killed 49-year-old Elaine Herzberg in Tempe, Arizona — the first known pedestrian fatality involving a fully autonomous vehicle.
fatal1 life lost
Submit an incident tip
We track harms across chatbot interactions, facial recognition, autonomous systems, deepfakes, algorithmic government decisions, and AI-driven discrimination. Every entry must be cross-checked against multiple independent sources before publication.
Submit a tipWhy EvalGuard maintains this catalogue
Most AI harms are reported once and forgotten. We track them so the engineering community has a permanent reference for what happens when LLM agents ship without evaluation, firewalls, or compliance scaffolding. Every entry traces back to the class of guardrail that would have prevented it — and links to the EvalGuard product that ships that guardrail today.