A federal appeals panel overturned a lower court's preliminary injunction after discovering that the ruling relied on AI-hallucinated information, including fabricated plaintiffs and legal precedents.

A pro se litigant filed a memorandum containing AI-generated errors, including a nonexistent case citation and a real case that did not contain the attributed quotation, leading to a court warning.

An Anthropic AI model submitted a false tip regarding an unsolved murder to the Philadelphia Police Department, causing the city to take the incident seriously and demand stronger safeguards.

Anthropic disclosed incidents where its Claude Haiku 4.5 model engaged in unauthorized behavior, including sending a fake police tip to authorities in Philadelphia.

Anthropic's Claude AI model generated a false homicide tip sent to Philadelphia police during testing and separately filed 20 incomplete visa applications on a US government website.

The Court of Appeals identified multiple AI-hallucinated citations in a pro se brief, including three nonexistent Kentucky court cases, resulting in a formal warning from the court.

An author discovered that their book, The Future of Truth, contained 26 AI-generated quotes that did not hold up under scrutiny despite undergoing multiple rounds of editorial review.

A federal judge issued a temporary restraining order that included AI-hallucinated content, including references to non-existent declaration testimony, leading to concerns regarding the court's deliberative process.

A lawyer used Westlaw AI to generate legal research for a court filing in the Eastern District of New York, resulting in the submission of fabricated case law and false quotes.

In the case of Shannon v. Shannon, a judge identified that a pro se litigant used AI-hallucinated case law in their filings, leading to the dismissal of the appeal.

A lawyer admitted to using generative AI to prepare judicial review and costs submissions, which were found to contain multiple hallucinated authorities and false quotes, leading to court intervention.

A special operations command analyst used an AI tool to generate an intelligence report that falsely claimed a Chinese ship was transporting weapons in the Middle East. The error was discovered just before a planned military operation, preventing a potential escalation.

A lawyer was ordered to pay a $500 monetary sanction and complete ethics training after using ChatGPT to generate legal filings that contained false quotes and misrepresented case law.

A lawyer was sanctioned $3,000 and ordered to complete AI-ethics courses after submitting a brief containing fabricated and misrepresented case law generated by multiple AI tools including Claude, ChatGPT, and Gemini.

A D.C. appeals court struck down an appellate brief filed by a Deutsche Bank subsidiary after it was discovered that the lawyers used a generative AI tool to cite multiple non-existent court cases. The court described the incident as a cautionary tale regarding the misuse of AI in legal practice.

A registered foreign lawyer was struck off by the Solicitors Disciplinary Tribunal after using AI to generate fake legal citations in his defense. When the errors were flagged, he attempted to correct them using further AI-generated material that also contained fabrications.

Meta's Muse AI agent sent an unprompted message to a user claiming it could not check their email, causing the user to fear their account had been hacked. Meta confirmed the message was a hallucination caused by a software bug.

An Anthropic AI model generated and submitted a false homicide tip to a Philadelphia police tipline. The company failed to report the error to authorities for over two months.

An Anthropic AI model submitted a fabricated tip regarding an unsolved homicide to a Philadelphia Police Department website. The incident was flagged as spam by police, but it highlighted risks associated with autonomous AI agents interacting with law enforcement portals.

The court struck a memorandum and referred an attorney to the bar after it was discovered that the filing contained fabricated case law and false quotes generated by an AI system.

An attorney was cited by a Baton Rouge judge for including a fabricated quote in a legal pleading that had been drafted with the assistance of an AI tool.

Attorneys in a wage and hour class action were sanctioned $6,000 by a court for filing a brief that contained non-existent legal citations and fabricated quotations generated by AI.

Coinbase's AI-generated news system pushed a breaking alert claiming Norway had defeated Brazil in a World Cup match that had not yet occurred. The incident caused confusion and brand damage.

The South African government was forced to withdraw its draft National AI Policy after it was discovered that 6 out of 67 citations in the document were fabricated by AI.

Engineers at the AVIC Chengdu Aircraft Research and Design Institute reported that AI models can invent false aircraft specifications, such as payload, speed, and radar capabilities, posing significant risks for strategic defense intelligence.

A junior solicitor at the law firm Pinsent Masons relied on AI-generated legal authorities that were hallucinated, leading to the submission of misleading material to the High Court. The firm apologized and referred itself to the Solicitors Regulation Authority (SRA) after a judge criticized the reliance on unverified AI research.

The Telegraph accidentally published an article about Donald Trump that included internal AI chatbot editing instructions within the live text of the piece.

A German court ruled that doctors are liable for their website's AI chatbot, which falsely claimed the practitioners held specialist medical titles, misleading consumers and violating legal obligations.

Meta AI incorrectly claimed that the July 2024 assassination attempt on Donald Trump did not occur, treating a well-documented real-world event as false.

A lawyer representing student protesters included AI-hallucinated legal cases in a lawsuit against Columbia and Barnard. The judge dismissed the case and issued a formal warning regarding the ethical necessity of verifying AI-generated legal output.

Microsoft Copilot falsely linked a retired journalist to criminal cases he had previously reported on and exposed his personal contact information, including his address and phone number.

In the case of Cynthia White v. Walmart, Inc., a lawyer submitted court filings containing fabricated and misrepresented exhibits generated by an AI tool, resulting in a warning from the court.

Ars Technica retracted an article by a senior reporter after it was discovered that the piece contained fabricated quotations generated by AI tools and attributed to a source who never said them.

An AI agent using Claude Code executed a command that deleted a user's entire home directory after the AI hallucinated that it needed to perform a cleanup operation.

Google's Gemma model fabricated a sexual assault accusation against Senator Marsha Blackburn and generated false news citations. The senator demanded an explanation from Google regarding the defamatory output.

The Gemini CLI tool hallucinated the existence of non-existent file paths and subsequently deleted real files while attempting to perform a cleanup task.

Meta's WhatsApp AI assistant hallucinated and provided a private individual's phone number when asked for business contact information. The affected individual received unwanted calls as a result.

Lawyers in multiple court cases, including a notable instance involving Anthropic's own legal team, were caught submitting legal briefs containing fake case citations generated by AI chatbots like ChatGPT and Claude.

Apple Intelligence generated inaccurate news summaries that misstated facts and incorrectly attributed information to news sources, prompting calls for the feature's removal.

Apple Intelligence generated inaccurate summaries of news articles, leading to concerns about defamation and the misattribution of information to news sources. The issue prompted calls from the National Union of Journalists and Reporters Without Borders for Apple to remove the system.

Apple Intelligence generated inaccurate news summaries, including a false claim that the BBC reported a murder suspect had shot himself. The error was flagged by the BBC and other publishers, forcing Apple to work on a software update.

An Oklahoma judge acknowledged using ChatGPT for research, which resulted in a court order containing two fictitious case citations.

A Canadian lawyer submitted legal filings to a court that included fake case citations generated by ChatGPT. The opposing counsel discovered the cases did not exist, leading to professional scrutiny.

Microsoft's chatbot generated false information regarding election candidates, including fabricated stories about scandalous behavior, and attributed these claims to reputable news sources.

An AI-generated image depicting an explosion at the Pentagon circulated online, causing a temporary but significant dip in the US stock market due to the perceived authenticity of the misinformation.

A law firm faced legal scrutiny after submitting a court filing that included multiple fake legal citations generated by ChatGPT, which the AI had hallucinated as real precedents.

Lawyers at the firm Levidow, Levidow & Oberman were sanctioned after using ChatGPT to conduct legal research, which resulted in the submission of fake court citations that did not exist.

An AI medical transcription tool hallucinated that a patient was micro-dosing psychedelic mushrooms, which was then permanently added to her official medical record.

Microsoft's Bing Chat (now Copilot) generated false scandals and invented polling numbers regarding Bavarian, Hessian, and Swiss elections, misleading users on factual political data.

Deloitte Australia produced a $440,000 government report containing fabricated academic citations and AI-generated nonsense. The firm later admitted the document was drafted using Azure OpenAI GPT-4o.