New York Times reporter John Carreyrou and five other writers sued xAI, Anthropic, Google, OpenAI, Meta and Perplexity in California federal court, accusing them of pirating their books to train AI models. Perplexity says it "doesn't index books." Pointedly shunning a class action, the writers argue such suits let defendants settle at "bargain-basement rates."
James Castell, 40, of Heathfield was given 18 months in prison, suspended for two years, at Lewes Crown Court after pleading guilty to possession and distribution charges. Sussex Police say he distributed an AI-generated indecent image of a child. They say his devices held more than 3,800 indecent images of children, plus AI software that generates images from text prompts.
According to 404 Media, Flock left livestreams and administrator control panels for at least 60 of its AI-enabled Condor cameras exposed to the open internet, without any password or login, letting anyone watch, download 30 days of archived video and change settings. Reporters watched cameras track pedestrians and film children on a playground, and one reporter watched themselves livestreamed in Bakersfield, California.
Instacart said Monday it was ending, effective immediately, AI-driven pricing experiments after a Consumer Reports and Groundwork Collaborative study of more than 400 shoppers found up to five different prices for the same item at the same store on the same day, with some paying 23% more. The company conceded the tests 'missed the mark for some customers.'
Andong's election commission decided to fine the maker of an AI-generated song praising Mayor Kwon Ki-chang, and a person who spread it in group chats, 5 million won each for omitting the required AI labelling, and reported the maker to police for alleged early campaigning. It is Korea's first such fine since deepfake election rules were added on 28 December 2023.
On December 22, 2025, the Delhi High Court restrained websites and platforms from using R. Madhavan's name or images commercially without consent, barred defendants from exploiting his personality traits through AI and deepfakes, and ordered certain obscene content removed. His counsel said one defendant made a fake 'Kesari 3' trailer and posted deepfake and AI-generated content as him.
A video posted on X by an account describing itself as a 'Parody AI Account' falsely showed Australian police announcing the arrest of four Indian Muslims over the December 14, 2025 Bondi Beach shooting. BOOM found it was a deepfake: the original briefing, published by The Guardian on December 18, made no such claim, and detection models flagged synthetic manipulation.
Who did it: HYBRID AI Hallucination Cases database
In an Australian visa judicial review, the Minister flagged that authorities cited in the applicant's submissions did not exist; on 26 May 2025 counsel admitted the representative had used an AI program and not verified the cases. The court sought an explanation and submissions on referral to South Australia's legal conduct commissioner, and dismissed the application on 22 December 2025, deferring the AI issues.
Who did it: HYBRID AI Hallucination Cases database
In a HABITAT trade mark dispute, Mimbim Enterprises told a delegate of Australia's Registrar of Trade Marks that early materials filed for Leytcorp by its representative, IP Wealth, cited several non-existent cases, alleged to be AI hallucinations. The delegate called the conduct 'unfortunate', attributed it to the representative rather than Leytcorp, and on 22 December 2025 granted Mimbim's extension requests with scale costs.
Pen Test Partners found that Eurostar's AI customer support chatbot validated only the latest messages in a conversation, letting earlier ones be altered to carry malicious prompts, and had unverified conversation and message IDs plus an HTML injection flaw enabling JavaScript in the chat window. Eurostar told City AM no sensitive customer data was at risk; the flaws have since been mitigated.
