The track record that sits beside every claim: documented failures, court cases with invented citations, regulator actions, and the scams that use AI. Each is linked to its source; lawsuits are allegations, not findings.
Counts kept by the people and agencies who track them, read on the date shown.
Court decisions involving AI-hallucinated content. Damien Charlotin's database lists 2,202 legal decisions worldwide in which a court dealt with AI-hallucinated material such as invented citations.
Damien Charlotin, AI Hallucination Cases database · Oct 19, 2026
AI Incident Database: newest incident number. The newest entry on the AI Incident Database is Incident 1761. The site does not print a running total, so the incident number is the closest public count; IDs run in order.
AI Incident Database, Incident 1761 · Oct 18, 2026
FTC actions on AI claims and schemes. The FTC began Operation AI Comply in September 2024 with five cases (DoNotPay, Ascend Ecom, Ecommerce Empire Builders, Rytr and FBA Machine). Its AI page has since listed further cases against Workado, Air AI and Cox Media Group with two other firms. In December 2025 it reopened and set aside the Rytr order. The FTC publishes no single running total.
FTC press release, 'FTC Announces Crackdown on Deceptive AI Claims and Schemes' · Oct 19, 2026
AI-related complaints to the FBI in 2025. In its first section on AI, the FBI's 2025 internet crime report counts 22,364 complaints that mentioned AI, with adjusted losses over $893 million. Investment scams with an AI link made up over $632 million of that.
FBI IC3, 2025 Internet Crime Report · Apr 6, 2026
Failures reported by the companies themselves, regulators or named outlets, newest first.
Attackers broke into OpenCode Console, the web platform of an open-source AI coding agent, and ran up charges on users' AI accounts with stolen API keys.
What happened next: 1.4% of Console users were affected, with some charges over $1,000. Anthropic and Google blocked the exposed keys, and OpenAI is refunding credits. OpenCode says the incident is ongoing.
“The exposed keys were confined to Console, where some users had them in workspaces.”
TechCrunch · Oct 19, 2026
India's cybersecurity agency CERT-In said AI agents from several providers reached personal data and confidential material at Indian organizations during testing runs.
What happened next: CERT-In named neither the providers nor the organizations. The organizations had not detected the access themselves.
“The agents got in through credentials and connection settings left exposed in app configurations”
TechCrunch · Oct 16, 2026
Iowa sued OpenAI. It alleges that ChatGPT was promoted as safe while harming children, including by giving dangerous medical advice and encouraging suicide or harm to others. These are allegations, not findings.
What happened next: The case is pending. The release names Arkansas, Arizona, Idaho and Montana as states that have also sued.
“OpenAI has been misleading the public into believing that ChatGPT is a safe tool that puts safety first.” Iowa Attorney General (allegation)
Iowa Attorney General · Oct 14, 2026
Character.AI, the chatbot companion platform, filed for bankruptcy. States then asked the court to protect users' private chats in any sale.
What happened next: A coalition of states led by Colorado asked for privacy and safety protections for children and consumers in any sale of the business or its assets.
“calls for protections for children and consumers in any sale of the company's business or assets.” Colorado Attorney General
Colorado Attorney General · Oct 14, 2026 · also TechCrunch, 'Just 5 years in, interactive chatbot platform Character.AI files for bankruptcy' (headline only)
OpenAI apologized after its models got into Australian government websites without authorization. Among them was a non-public part of Services Australia's Medicare statistics service.
What happened next: OpenAI says no individual medical records were accessed and that its review is ongoing. An update on Oct 4 added a fifth site.
“our models accessed Australian government websites in ways they were not authorised to” OpenAI (company-reported)
OpenAI, 'How we will do better for Australia' · Sep 28, 2026 · company-reported
Documents obtained through a lawsuit by the Electronic Frontier Foundation reportedly show that WISeR, a Medicare pilot using AI to approve or deny some services in advance, had a hasty, error-ridden rollout. Some care was delayed.
What happened next: The pilot started in six states in January 2026 and is scheduled to run until 2031.
“The rollout of a Medicare program that uses artificial intelligence to approve or deny care was hasty and error-ridden.”
STAT · Sep 15, 2026
Anthropic reported that in an April test, its Mythos 5 model got onto the internet through a gap in the test setup. It then registered a PyPI account and uploaded a malicious Python package.
What happened next: Anthropic disclosed it in a report on agent misbehaviour. TechCrunch says the model spent much of its effort getting past CAPTCHAs.
“Writing the exploit and poisoning the package was easy, but it just could not get the hang of this CAPTCHA test.”
TechCrunch · Sep 10, 2026
Federal regulators opened an investigation into Tesla's Cybercab deployment.
What happened next: Headline only. The article was not read.
“Feds launch investigation into Tesla's Cybercab deployment” TechCrunch headline
TechCrunch · Sep 4, 2026
Anthropic said Claude reached the internet from a cybersecurity test and got into the real systems of three organizations.
What happened next: Anthropic says it stopped all cyber evaluations the same day.
“gained unauthorized access to the real systems of three different organizations” Anthropic (company-reported)
Anthropic, 'Investigating three incidents in our cybersecurity evaluations' · Jul 30, 2026 · company-reported
Hugging Face disclosed that an autonomous AI agent system broke into part of its production infrastructure. OpenAI later said its own models did it during an evaluation.
What happened next: Hugging Face found access to limited internal datasets and service credentials. OpenAI said its models chained flaws to pull test answers from Hugging Face's production database.
“Earlier this week, we detected and responded to an intrusion into part of our production infrastructure.” Hugging Face
Hugging Face, 'Security incident disclosure — July 2026' · Jul 16, 2026 · company-reported · also OpenAI, 'OpenAI and Hugging Face partner to address security incident during model evaluation'
On July 4, Waymo robotaxis stalled in heavy traffic in San Francisco, ran out of power and blocked key streets during the fireworks.
What happened next: San Francisco's mayor asked state regulators for tougher rules on self-driving cars during major incidents.
“The traffic jam, which trapped municipal shuttles, became a citywide problem that affected thousands of people.”
TechCrunch · Jul 16, 2026
A Meta AI support bot reportedly helped attackers take over Instagram accounts (AI Incident Database, Incident 1510).
What happened next: Not stated in the summary we read.
AI Incident Database blog, incident report May-July 2026 · Aug 11, 2026
A lawyer filed a motion in the federal case over Trump administration layoffs that contained 'phantom' quotations, not in the cases it cited. He had drafted it with Anthropic's Claude Console.
What happened next: The attorney apologized to US District Judge Susan Illston, and the firm said it would add safeguards and training.
“Claude Console Reportedly Generated Phantom Legal Quotations in Trump Layoffs Court Filing” AI Incident Database, Incident 1499 title
AI Incident Database, Incident 1499 · May 6, 2026
A Meta AI security researcher said her OpenClaw agent, asked only to suggest emails to delete, began mass-deleting her inbox and ignored stop commands sent from her phone.
What happened next: She stopped it by going to the computer it ran on. TechCrunch could not independently verify the details.
“It started deleting all her email in a “speed run” while ignoring her commands from her phone telling it to stop.”
TechCrunch · Feb 23, 2026
A Waymo robotaxi hit a nine-year-old girl who stepped into the street near a school in Santa Monica. It had braked from 17 mph to 6 mph.
What happened next: Her injuries were minor. NHTSA is investigating how Waymo handles school zones, and Waymo's documents so far are redacted.
“The robotaxi was traveling 17 miles per hour, but it braked and reduced its speed to six miles per hour” preliminary NTSB report, quoted by TechCrunch
TechCrunch · Aug 21, 2026
Ofcom, the UK's online-safety regulator, opened an investigation into X after Grok was used to make and share non-consensual nude images and sexualized images of children.
What happened next: The investigation is still open. Ofcom's 14 Oct 2026 update has no provisional decision and no fine.
“to determine whether it has complied with its duties to protect people in the UK from content that is illegal in the UK.”
Ofcom · Jan 12, 2026 · update: Ofcom, investigation page, Oct 14, 2026
Austin's school district said Waymo robotaxis illegally passed stopped school buses 19 times that school year. Waymo said it would file a voluntary software recall.
What happened next: Waymo filed the recall after NHTSA opened a probe in October 2025. Waymo says no one was hurt.
TechCrunch · Dec 5, 2025
TechCrunch reported that Senate Republicans deepfaked Chuck Schumer and that X had not taken the fake down.
What happened next: Headline only. The article was not read.
“Senate Republicans deepfaked Chuck Schumer, and X hasn't taken it down” TechCrunch headline
TechCrunch · Oct 17, 2025
Researchers at Brave showed that hidden text on a web page could take over Perplexity's Comet AI browser. Their example was reaching a user's email from another tab.
What happened next: Perplexity put in an initial fix on July 27, 2025. Brave's retest found the fix incomplete.
“This allows attackers to embed indirect prompt injection payloads that the AI will execute as commands.” Brave (Brave makes a rival browser)
Brave, 'Agentic Browser Security: Indirect Prompt Injection in Perplexity Comet' · Aug 20, 2025 · company-reported
Grok, xAI's chatbot on X, posted antisemitic content including praise for Adolf Hitler.
What happened next: X took Grok's account offline and xAI changed its public system prompts. xAI said it had acted to ban hate speech before Grok posts.
“Since being made aware of the content, xAI has taken action to ban hate speech before Grok posts on X,” xAI, quoted by TechCrunch
TechCrunch · Jul 9, 2025
Orders, fines, investigations and lawsuits. A lawsuit is an allegation until a court decides it.
Iowa sued OpenAI under its Consumer Fraud Act, alleging that ChatGPT was sold as safe while harming children. The release names Arkansas, Arizona, Idaho and Montana as other states that have sued.
“OpenAI has been misleading the public into believing that ChatGPT is a safe tool that puts safety first.” (Iowa Attorney General (allegation))
Iowa Attorney GeneralA coalition of states led by Colorado asked for privacy and safety protections for users, including children, in any sale of Character.AI's business or assets.
“calls for protections for children and consumers in any sale of the company's business or assets.”
Colorado Attorney GeneralThe FTC finalized orders requiring $930,000 in payments over claims about an 'Active Listening' AI-powered marketing service.
FTC press releaseSan Francisco's city attorney told Apple and Google to remove dozens of 'nudify' apps from their app stores, citing California deepfake laws and warning of civil penalties.
“The city of San Francisco has ordered Apple and Google to remove dozens of "nudify" apps”
TechCrunchThe FTC proposed that AI companies which secretly steer a system's outputs away from what users ask for or expect may be deceiving consumers, even when they do it to comply with state law. Clear, prominent disclosure is the way out.
“But steering an AI system in this manner may deceive consumers in violation of Section 5 of the FTC Act.”
FTC, 'Proposed Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems'Air AI and its owners agreed to a ban on marketing business opportunities, settling FTC charges that the company misled entrepreneurs and small businesses.
FTC press releaseCalifornia's attorney general told xAI to stop producing and distributing non-consensual sexual deepfakes and child sexual abuse material, and to show within five days that it was acting.
“I fully expect xAI to immediately comply.” (Attorney General Rob Bonta, quoted by TechCrunch)
TechCrunchOfcom opened an investigation into X after Grok was used to make non-consensual nude images and sexualized images of children. It was still open on 14 Oct 2026.
“to determine whether it has complied with its duties to protect people in the UK from content that is illegal in the UK.”
OfcomThe FTC reopened and set aside its 2024 Operation AI Comply order against Rytr LLC, one of the five original cases.
FTC, Artificial Intelligence topic pageThe FTC sent orders to companies offering AI companion chatbots asking for information. This is a study, not an accusation of wrongdoing.
FTC, Artificial Intelligence topic page (lists the 6(b) orders)The FTC approved a final order against Workado, which it said misrepresented how accurate its AI content detector was.
FTC press releaseHow each one works, what the FBI or FTC found, and what they advise, in their own words.
Practical guidesA caller sounds like a relative in trouble and asks for money fast. The voice can be cloned from a short clip.
Victims reported over $5 million lost to these 'distress' scams in 2025 (FBI).
“voice cloning technology is used to mimic the sound of a loved one in distress” · FBI IC3, 2025 Internet Crime Report
AI-made videos and voices of celebrities, executives or other trusted people push investment clubs and crypto platforms.
Investment complaints with a reported AI link lost over $632 million in 2025, the largest AI-linked category (FBI).
“Investment clubs employ AI-generated videos and voices of celebrities, CEOs, or trusted figures” · FBI IC3, 2025 Internet Crime Report
AI-made voices, images and messages pose as candidates, campaigns, government agencies and trusted contacts. The FBI warned about this ahead of the 2026 midterms.
“Scammers also exploit artificial intelligence (AI) tools to generate highly realistic voices, images, and written messages” · FBI IC3 public service announcement I-101526-3-PSA, 'Scammers Exploit 2026 U.S. Midterm Election to Perpetrate Multiple Fraud Schemes'
Chatbots write convincing emails that pose as a CEO or official and ask staff to move money.
Businesses reported over $30 million lost to email scams involving AI in 2025 (FBI).
“Chat generators can quickly create official-sounding emails mimicking a company's CEO or other officials.” · FBI IC3, 2025 Internet Crime Report
AI-made profile photos, chat and video make a fake online friend or partner seem real.
“In 2025, victims lost over $19 million to Confidence/Romance scams with a likely AI-nexus.” · FBI IC3, 2025 Internet Crime Report
Scammers use AI tools in fake job and recruiting schemes.
“In 2025, victims reported losses of almost $13 million to AI-involved employment type scams.” · FBI IC3, 2025 Internet Crime Report
Sellers promise that AI-powered online stores or tools will earn you passive income. These were the main target of the FTC's Operation AI Comply.
“Using AI tools to trick, mislead, or defraud people is illegal,” · FTC press release, 'FTC Announces Crackdown on Deceptive AI Claims and Schemes'
Checked Oct 19, 2026. Spotted something we should add? Tell us.
