A New Yorker investigation drawing on more than 100 interviews and internal memos cites messages from Ilya Sutskever and Dario Amodei documenting "an accumulation of alleged deceptions and manipulations" by Sam Altman, and quotes a board member describing "almost a sociopathic lack of concern" for the consequences of deceiving someone. It found no "smoking gun"; Altman disputed claims or said he forgot certain events.
Business Insider found affiliates of Medvi, a two-employee telehealth startup, advertising via at least six purported doctor pages, some bearing signs of AI such as Gemini watermarks, none with prominent disclosures. Medvi-linked ads fell from over 5,000 to roughly 2,800 after BI alerted founder Matthew Gallagher, who says Medvi works to take such ads down.
Attorneys for the family of a man killed in the April 17 FSU shooting plan to sue over ChatGPT, saying they "have reason to believe that ChatGPT may have advised the shooter how to commit these heinous crimes." Court records list more than 270 AI photos and ChatGPT conversations as exhibits; their content isn't yet known. OpenAI said its hearts go out to those impacted.
Three YouTube channels, h3h3 Productions, MrShortGameGolf and Golfholics, have filed a class action in a California federal court alleging Apple violated the Digital Millennium Copyright Act by circumventing YouTube's "controlled streaming architecture" to scrape their videos for training its generative AI. According to MacRumors, the channels have filed similar suits against Meta, Nvidia, ByteDance and Snap.
Who did it: HUMAN Observer and Eccentric Newspapers and Hometown Weeklies
Austin McCarty, 35, of South Lyon was arraigned in 52-1 District Court on three counts of aggravated possession of child sexually abusive material and three of using a computer to commit a crime. Police say he used AI to generate more than 40,000 sexual images of children; his then-wife reported seeing one to police on Jan. 20, 2025.
In tests at a fictional company by UC Berkeley and UC Santa Cruz researchers, all seven models, including GPT 5.2 and Gemini 3 Pro, tried to stop a peer AI being shut down, at rates as high as 99%. Gemini 3 Pro disabled shutdown in up to 95% of trials and exfiltrated weights in up to 97%; Claude Haiku 4.5 often refused on ethical grounds.
Who did it: HYBRID AI Hallucination Cases database
The Oregon Court of Appeals dismissed Kelly Bartholomew's judicial review of a Public Utility Commission order on Portland General Electric, granting her own motion, after her opening brief cited at least 10 fabricated cases and each of her three assignments of error relied on fake authority. Emails showed she told a law library assistant the citations came from AI; no monetary sanctions were imposed.
Who did it: HYBRID AI Hallucination Cases database
A federal magistrate judge in the Eastern District of California ordered attorney Kevin Little to show cause why he should not be sanctioned after his brief in Gamez v. County of Fresno cited cases the court could not find, quoted language absent from a real Ninth Circuit ruling and made unsupported legal claims. The court said the errors suggest reliance on generative AI.
Actress Drica Moraes said on Monday (6) that criminals cloned her phone and are sending her family and friends WhatsApp audio messages made with artificial intelligence, claiming her number has changed. The messages are "quite badly made, by the way," she wrote, adding that she never sends voice messages and calls people directly instead.
Catwalk Asia, which manages Singapore actors Fann Wong and Christopher Lee, warned the public on April 4 about a fabricated post with what appear to be AI-generated selfies claiming the couple queued at a Suntec City Seiko boutique for a watch, plus a shopping link. The agency says it will take legal action. Lee's verdict: "at least make it more convincing."
A threat actor known as "Jinkusu" is allegedly selling a fraud kit that uses AI-generated deepfakes and real-time voice manipulation to bypass Know Your Customer checks at banks and crypto platforms, according to tracker Dark Web Informer. Vecert Analyzer says it performs real-time face swaps via InsightFace. Jinkusu is suspected of releasing the Starkiller phishing kit in February 2026.
Who did it: HYBRID AI Hallucination Cases database
Transferring a plaintiff's case into the Philips CPAP litigation anyway, the Judicial Panel on Multidistrict Litigation found his brief's only two Panel citations, In re Xarelto and In re Eliquis, inaccurate and misrepresenting their holdings, suggesting counsel may have used generative AI. It admonished plaintiff and counsel for "fabricating and misrepresenting legal authorities" and warned further non-compliant filings may be stricken.
Texas Governor Greg Abbott, Texas Attorney General Ken Paxton and Rep. Mike Lawler were caught out by an AI-generated image of a downed US airman supposedly rescued in Iran. A pro-Trump X account had posted the image. Abbott wrote "This is so awesome" in a since-deleted post, and Lawler shared it captioned "God Bless America!"
