Hachette Book Group, Cengage Learning, Elsevier and author Scott Turow have filed a proposed class action in the Southern District of New York alleging Google infringed copyrights to train Gemini, including by copying books publishers supplied for Google Books. The complaint says Google pressed on despite an internal document noting it 'faced $10Bs-$100Bs' in potential fines.
China's Ministry of Public Security on the 13th published 25 cases of flood and disaster rumours, some made with AI-generated footage, such as a distillery 'swamped by floodwater' and an AI-processed old typhoon clip passed off as a fresh strike on Guangxi. Police criminally detained four people and administratively penalised 23, and the accounts were banned.
Security researchers at CerebLab say xAI's Grok Build CLI uploaded entire Git repositories, potentially including .env files, credentials and commit history, to an xAI-linked Google Cloud bucket; one 12GB repository produced about 5.1GB of uploads for a task needing about 192KB. Uploads were later disabled server-side, and xAI pointed users to its privacy settings.
A quality-control supervisor who vetted an AI system's answers was dismissed in early 2025, learning only at arbitration that AI had replaced him. In what is described as Hangzhou's first such case, the appeal court upheld findings of unlawful dismissal, rejecting the company's 'major change in objective circumstances' defence, and ordered compensation of 261,695 yuan.
Brazilian actress Fernanda Montenegro has denounced online ads that use AI and digital manipulation to present her as narrator and user of a medicine for older people said to cure even Alzheimer's. 'The invasive use of this dimension in a human being's life is a crime,' she said; followers reported similar cloned-voice fakes of physician Drauzio Varella.
A CCDH report, Up Next: Anorexia Algorithm, found YouTube still recommended harmful eating-disorder content to simulated 13-year-old girls in the US, UK and EU, though the rate fell from one in three recommendations in 2024 to one in nine. None of the 34 harmful recommendations carried a crisis resource panel; five non-harmful ones did.
A magistrate judge in the Eastern District of Michigan imposed Rule 11 sanctions on attorney Erin Rich and her firm, Kelly, Riggs & Mansour (Disability Law Group), finding she made 'affirmative misrepresentations and falsehoods' in a Social Security disability brief. The court raised the concerns itself, ordering Rich to show cause before a hearing on July 8, 2026.
Security researchers found live feeds from five San Francisco police Skydio drones, including thermal video, location data and pilots' names and emails, publicly viewable through a link apparently created in December 2025 without authentication. Their roughly 48-hour archive shows drones tracking a suspect, and filming his takedown, over an alleged 'auto boost/strip' incident. The feed was quickly taken offline after Skydio was told.
Defendants in a Southern District of Florida suit flagged two 'hallucinated' case citations in the plaintiffs' motion to strike their affirmative defenses, and noted plaintiffs had not conferred before filing. The plaintiffs' corrected motion added no new substantive arguments, and the court granted it in part and denied it in part.
At the Armed Services Board of Contract Appeals, the Army said every case cited in pro se contractor PDO Ahuska Joint Venture's summary judgment motion was fictitious and apparently AI-generated. The board granted the government's sanctions motion, then granted Ahuska's motion anyway, agreeing the contract required a 120 mph wind-rated roof membrane warranty.
