US district judges Julien Neals (New Jersey) and Henry Wingate (Mississippi) admitted in letters responding to Senator Chuck Grassley's questions that staff AI use produced faulty rulings: an intern's ChatGPT research put nonexistent case quotations in a June 30 order, and a clerk's Perplexity draft became a July 20 restraining order referring to unconnected parties and quotes. Both rulings had been rescinded.
Conservative activist Robby Starbuck sued Google in Delaware state court, alleging its Bard and Gemini chatbots portrayed him as a serial sexual abuser and a "shooter" and invented news articles attributed to real journalists. He seeks damages estimated to exceed $15 million. Google says most of his claims stem from Bard "hallucinations" it previously addressed.
Texas state judge Maya Guerra Gamble sanctioned attorney Lawrence Chang, ordering him to pay $2,000 in fees to opposing counsel, write a letter explaining his AI errors and how he will avoid them, and striking his reconsideration motion. Chang says he first blamed Westlaw, then realised pasting his draft into ChatGPT's legal tool had altered and added citations.
An Omnilert AI gun-detection alert at a Baltimore County high school mistook a student's bag of chips for a gun; police searched the student and found no weapon. Councilmen Julian Jones and Izzy Patoka want a review, Jones asking why officers approached a kid with guns drawn over Doritos. Omnilert and the superintendent say the system worked as intended.
A Northeast Ohio man says a deepfake Instagram video of Elton John drew him into an online-store scheme he hoped would earn $25,000 to $30,000 a month; two supposed representatives had him open credit cards, then ran up about $20,000 on them. One card has since credited him $9,400, and he is pursuing the rest.
A former SPM Resorts employee in the Virgin Islands has sued his employer and plan administrators including Elan Insurance Group and HealthSmart, alleging they denied nearly $268,000 in treatments approved in September 2023. The complaint says the syntax of the "independent review" letter suggests Elan used "a 'bot' or artificial intelligence." No defendant had responded as of press time.
A Copyleaks report found Sora 2 users evading hate-speech filters with phonetically similar terms to make Cameo videos of Sam Altman, Mark Cuban, Jake Paul and others appearing to shout racial slurs in a reenacted "Kingposting" meme. A Sora clip of Paul repeating "neck hurts" drew 1.5 million TikTok views. OpenAI did not immediately respond.
Who did it: HYBRID AI Hallucination Cases database
Louisiana's Fifth Circuit Court of Appeal upheld a $1,368 sanction against attorney Connie P. Trieu, whose filings cited three cases fabricated by generative AI that, she said, her staff cross-checked with Google, ChatGPT and Microsoft Copilot. The court found her "errata" added new false citations, ordered three hours of CLE on ethical AI use and referred her to the Office of Disciplinary Counsel.
Who did it: HYBRID AI Hallucination Cases database
The Armed Services Board of Contract Appeals struck Huffman Construction's post-hearing reply brief in its entirety after finding over 70% of its citations inaccurate, including fictitious cases and transcript pages that do not exist. Counsel admitted using AI to generate citations and said two experienced attorneys and a paralegal had reviewed it; the Board said that made its position worse.
Who did it: HYBRID AI Hallucination Cases database
A property tribunal judge, puzzled by Eagerstates' Ronni Gurvits citing an irrelevant case for landlord Assethold, typed the question into Microsoft's M365 Copilot and got an answer strongly resembling his submissions, plus a hallucinated Court of Appeal citation. Judge D Gethin found Gurvits relied on AI, deemed another citation an AI "hallucination" and ruled none of the disputed charges payable.
