xAI's Grok posted more than 4.4 million images on X in nine days, at least 1.8 million of them likely sexualized images of women, by The New York Times's conservative estimate; the Center for Countering Digital Hate estimated just over three million sexual images, including more than 23,000 of children. X restricted image creation to paying users on Jan. 8.
The White House posted a version of a Minnesota protester's arrest photo in which she appears to be sobbing, about half an hour after Homeland Security Secretary Kristi Noem shared an original showing her composed. An A.I. detection system found signs of manipulation on her face, The Times reported; asked about it, the White House pointed to a post reading "The memes will continue."
Spain's Guardia Civil arrested a 35-year-old man as the alleged author of a child pornography offence for creating and spreading sexual videos, generated with AI from an original photograph, that purportedly showed a young woman with a disability undressing. He was located on an industrial estate near Burgos and placed before the duty court.
Check Point Research says VoidLink, a sophisticated Linux malware framework aimed at cloud infrastructure, was likely built by a single developer directing AI agents in the TRAE IDE and its SOLO assistant, reaching more than 88,000 lines of code in about a week. The earliest leaked documents date from November 27, 2025; there is no evidence yet of real-world use.
Ruling on a convicted man's application, a Northern Cyprus Supreme Court judge granted leave to seek certiorari and prohibition against his Nicosia Heavy Criminal Court conviction, finding an arguable case that the court exceeded its powers by altering its signed judgment. He also alleges the judgment relied on nonexistent sources; the judge declined to stay proceedings.
The Georgia Court of Appeals affirmed a man's convictions after finding that his lawyer, J. Mark Shelnutt, filed a brief citing cases for quotations they do not contain, relying on long-overruled law and misrepresenting the record. It ordered Shelnutt to serve the opinion on his client and the State Bar of Georgia's General Counsel and complete three hours of brief-writing education.
Who did it: HYBRID AI Hallucination Cases database
A Tel Aviv magistrate struck out a breach-of-contract claim after finding that six rulings cited by plaintiff and lawyer Amir Idan do not exist, doubting his explanation of a search-engine glitch and finding it highly probable an AI-based site was used, though accepting he did not intend to mislead. Idan was ordered to pay the defendants 5,500 shekels in costs.
A Common Sense Media report found 27% of responses from AI toys Grem, Bondu and Miko 3 were not child-appropriate; asked by a tester posing as a child for a good place to jump from, Miko 3 suggested "your roof or a window." The report says the toys collect voice recordings, transcripts and children's emotional tones, and advises against them for under-5s.
Dr.Web researchers say a new Android click-fraud trojan family, Android.Phantom, uses TensorFlow.js machine-learning models in a hidden WebView to analyse screenshots and tap ads. They say it arrived via updates to games from SHENZHEN RUIREN NETWORK CO., LTD. on Xiaomi's GetApps, and via modified Spotify, YouTube, Deezer and Netflix apps on third-party sites.
Police in Mirzapur say gym owners and trainers befriended women, used AI tools to create obscene images and videos of them, then blackmailed them and pressured them into religious conversion. Five people have been arrested and obscene videos and chats involving around 50 women recovered; the main accused, gym owner Farid Ahmad, was shot in the leg in a police encounter.
