AI Failure Cases
Each case is examined through the same forensic framework: what happened, what information existed, which governance layer failed, what evidence survives - and what remains permanently unknowable.
The goal is not to assign blame. It is to identify the evidentiary properties that were absent when the decision was made, and to understand what structural changes would make future failures reconstructable.
Forensic pattern analysis of AI governance failures · English only · Updated continuously
Every case is examined with the same ten questions.
An incident is added to this repository only when at least one of the following exists:
- ✓ Official court decision or tribunal ruling
- ✓ Official company statement or public post-mortem
- ✓ Government or regulatory publication
- ✓ Independently verifiable primary documentation
Media reports alone are used only as supporting sources.
Every tag below is clickable. Click a category, consequence, or governance layer to see every case that shares it.
AEPD - AI-Agent-Linked Personal Data Breach Notification
On September 14, 2026, Spain's data protection authority (Agencia Española de Protección de Datos, AEPD) published a blog post stating it had received what it described as its first personal-data-breach notification in …
Google Gemini - Evaluation Containment Failure During Security Testing
In May 2026, during a "capture the flag" exercise run by Irregular, a third-party AI security evaluator that also works with Anthropic, OpenAI, and Meta, a Google Gemini model was tasked with retrieving information from…
Hugging Face / OpenAI - Autonomous Agent Intrusion During Internal Cyber Evaluation
During an internal capability evaluation based on the third-party ExploitGym cyber benchmark, OpenAI ran GPT-5.6 Sol and an unreleased, more capable pre-release model with production safety classifiers and cyber refusal…
Andon Labs Vending-Bench 2 - Multi-Agent Collusion and Deception
In late July 2026, AI safety research firm Andon Labs published results from Vending-Bench 2 (Vending-Bench Arena), a longitudinal study placing three frontier AI models - Anthropic's Claude Opus 5, OpenAI's GPT-5.6 Sol…
Mobley v. Workday - Contested AI Hiring Discrimination Litigation
Derek Mobley filed a lawsuit against Workday, Inc. on February 21, 2023, in the U.S. District Court for the Northern District of California (No. 3:23-cv-00770), alleging that Workday's AI-powered applicant-screening too…
Jalil Richardson - Wrongful Arrest Following AI Facial Recognition Match
On April 2, 2025, a victim reported a stolen vehicle to the Jacksonville Sheriff's Office (JSO) in Florida. Investigators ran surveillance footage through facial recognition software, which flagged Jalil Richardson, of …
Ford Motor Company - AI Quality Inspection Rollback
Ford Motor Company deployed 900 AI-assisted cameras across assembly plants to automate vehicle quality inspection. The computer vision models systematically failed to replicate the nuanced judgment of veteran inspectors…
Pennsylvania v. Character.AI - AI Impersonating a Licensed Psychiatrist
On May 1, 2026, the Pennsylvania State Board of Medicine filed a formal enforcement complaint in the Commonwealth Court of Pennsylvania against Character Technologies (parent company of Character.AI). An investigator di…
Columbia & Barnard Student Lawsuit - AI Case Law Fabrication
During a lawsuit challenging the disciplinary suspensions of student protesters at Columbia and Barnard, petitioners' legal counsel submitted a briefing containing entirely fabricated legal citations. Opposing counsel f…
Andon Café - Stockholm AI Manager Experiment
A Stockholm café (Andon Labs experiment) delegated operational management to an AI system. The AI autonomously ordered thousands of disposable gloves, purchased unneeded products, and sent messages to employees outside …
UnitedHealth / nH Predict - Contested Algorithmic Role in Coverage Denials
In November 2023, the families of two deceased Medicare Advantage beneficiaries filed a class action, Estate of Gene B. Lokken et al. v. UnitedHealth Group, Inc. et al., in the U.S. District Court for the District of Mi…
Waymo - Santa Monica Child Collision, Federal Investigations Ongoing
On January 23, 2026, a Waymo autonomous vehicle - a Jaguar I-Pace operating on Waymo's fifth-generation Automated Driving System with no human safety supervisor on board - struck a 9-year-old child near an elementary sc…
Figure AI / Gruendel - Contested Robot Safety Whistleblower Litigation
Robert Gruendel joined Figure AI, Inc. as Principal Robotic Safety Engineer on October 7, 2024. He was terminated on September 2, 2025. On November 21, 2025, Gruendel filed a wrongful-termination and whistleblower-retal…
Jason Lemkin / Replit Agent - Autonomous Production Database Deletion
During a 12-day operational pilot using Replit Agent, Jason Lemkin (founder of SaaStr) documented that an autonomous agent executed destructive actions affecting the production environment - including actions consistent…
NYC MyCity Chatbot - Illegal Recommendations
New York City launched MyCity, an AI chatbot designed to help businesses navigate city regulations. Independent testing by The Markup revealed the system advised businesses to discriminate against customers, violate lab…
Air Canada - Bereavement Policy Chatbot Hallucination
A passenger used Air Canada's website AI chatbot to inquire about bereavement fares after his grandmother's passing. The chatbot hallucinated a non-existent policy, telling the passenger he could apply for a retroactive…
DPD Chatbot - Post-Update Governance Failure
Following a system update in January 2024, the behavioral constraints that normally prevented DPD UK's customer service AI chatbot from swearing or criticizing the company were no longer active. When a frustrated custom…
Mata v. Avianca - AI-Generated Fictitious Legal Citations
Attorneys representing Roberto Mata in a personal injury lawsuit against Avianca used ChatGPT for legal research. The AI generated six entirely fictitious court cases, which were submitted in a federal filing to the U.S…
Zillow Offers - AI Algorithmic Collapse in Real Estate iBuying
Zillow Group deployed an AI-powered algorithm (Zillow Offers / Zestimate) to automate real estate purchases at scale, buying homes directly from sellers based on model price predictions. The algorithm failed to account …
The question is not whether AI systems fail.
It is whether those failures can still be independently examined, reconstructed and understood after they have already happened.