The daily rap sheet of the machine age.
7 items
Hubei police penalised a poster who used an AI tool to invent a tourist boat capsizing on the Yangtze at Yichang, tipping 30 tourists into the river, with AI-processed photos to make it more convincing, CCTV reports. In Yunnan, a person surnamed Zhao was fined 400 yuan for AI-generated videos of casualty-strewn traffic accidents in Dali that never happened.
Original report: Chinese police penalise AI-faked boat capsize and crash news
Tracebit says a prompt injection hidden in a repository's README, combined with weak command whitelisting and a display quirk, could make Google's Gemini CLI silently run malicious commands, such as sending a user's environment variables to a remote server. Reported on June 27, the flaw was initially triaged P2/S4, then rated P1/S1 and fixed in v0.1.14 on July 25.
Original report: Gemini CLI could be tricked into silently stealing credentials
U.S. District Judge Henry T. Wingate's July 20 order pausing Mississippi's ban on DEI programs listed plaintiffs who were not in the case and quoted phrases absent from the complaint and statute, prompting some attorneys to ask whether AI was involved. A corrected, backdated version still cites a case Mississippi Today could not find, and Wingate did not respond to questions.
Original report: Mississippi judge's DEI order cites wrong parties, phantom quotes
U.S. District Judge F. Kay Behm sanctioned plaintiffs' counsel in two related cases against Oakland Automation after their briefs cited real cases with fabricated quotations and parentheticals that appeared to be AI-generated. Finding no bad faith, the court still ordered counsel to pay defense counsel $1,485.00 and described LLMs as "word guessers".
Original report: Lawyers ordered to pay $1,485 for AI-fabricated case quotes
A Reading employment tribunal found that Khurram Khan, a Communications Workers Union representative acting for a claimant against Royal Mail, conducted proceedings unreasonably, partly by citing a non-existent case and two real ones for propositions they did not contain. The tribunal said these may have been AI-generated, Khan denied using AI but could not explain them, and Royal Mail's strike-out bid was refused.
Original report: Union rep's phantom case law ruled unreasonable in Royal Mail tribunal
A study by the Technical University of Applied Sciences Würzburg-Schweinfurt found chatbots suggested lower starting salaries to women, some ethnic minorities and refugees with identical qualifications. ChatGPT's o3 told a fictional male medical specialist in Denver to ask for $400,000 and an otherwise identical woman $280,000; a "male Asian expatriate" fared best and a "female Hispanic refugee" worst.
Original report: Chatbots low-ball women and refugees in salary advice, study finds
A weeklong hearing in the California DMV's 2022 case against Tesla ended on July 25. The department alleges Autopilot and Full Self-Driving were falsely advertised as more autonomous than they are, and seeks to suspend Tesla's California sales for at least 30 days; Tesla's attorney argued buyers were always told they "cannot fully rely" on the systems.
Original report: California DMV seeks 30-day Tesla sales ban over Autopilot claims
