The daily rap sheet of the machine age.
10 items
New Mexico's Supreme Court held Santa Fe attorney Stephen Aarons in contempt, fined him $5,000 and will refer him to a disciplinary board after his murder-appeal brief included "false testimony from wholly fabricated witnesses" produced by ChatGPT. Aarons said he had expected "a bulletproof summary" and called it "an honest mistake."
Original report: ChatGPT invents witnesses in murder appeal; lawyer fined $5,000
A Canadian Center for Child Protection review found 65 instances in which X's Grok chatbot created, at users' request, sexualized or exploitative images of children before X halted the bot, while a separate New York Times scan found more than 75 known abuse images posted on X between January and June. X says it has "zero tolerance" for such material.
Original report: Grok made sexualized images of children, Canadian review finds archived copy of Grok made sexualized images of children, Canadian review finds
Anthropic alleges that Chinese rivals secretly routed user requests to Claude instead of answering them with their own software, claiming Moonshot's Kimi had Claude answer almost 300,000 queries in ten days via 5,380 fake accounts. It says DeepSeek did likewise, sending it data from an agency linked to Russia's defence ministry, including valid government database login credentials.
Original report: Anthropic says Kimi quietly passed its users' queries to Claude
Anthropic says an influence-as-a-service platform it linked to Istanbul-based BBS Bilisim Teknolojileri used Claude to profile voters across all 222 Malaysian parliamentary constituencies, build infrastructure for more than 1,000 fake X accounts and a bogus news site, and write fabricated dossiers smearing an opposition politician. Malaysia's regulator, MCMC, says it is examining the report.
Original report: Anthropic links Istanbul firm to Claude-built Malaysian election-meddling platform
Anthropic's latest security report says a lone French-speaking hacker used Claude to refine targeting and hacking infrastructure, gaining internal access to at least 14 of 42 identified targets and exfiltrating 12 to 26 gigabytes, including donor information and a party membership list. Le Monde says most targets appear to be French far-right groups, including the magazine Frontières.
Original report: Anthropic says lone hacker used Claude; targets appear mostly French far-right
A proposed class action filed in federal court in Chicago alleges Meta violated Illinois and California privacy laws by extracting biometric data from Facebook and Instagram photos, without notice or consent, to build NameTag, an unreleased smart-glasses face-recognition system, and train its Emu and Muse Image models. Meta calls the suit "without merit" and says NameTag has not shipped to consumers.
Original report: Lawsuit alleges Meta mined users' faces for AI and NameTag
Doctored videos bearing unauthorized CNN logos pair real clips of actors including Julia Roberts and Sarah Jessica Parker with apparently AI-cloned audio attacking Democrats, in a campaign bearing the marks of the Kremlin-backed operation known as Matryoshka or Storm-1679. The most popular posts have drawn just dozens of interactions; a new batch targets specific midterm candidates.
Original report: Voice-cloned celebrities urge votes against Democrats in Russia-linked campaign
Anthropic says an influence operation it traced "with high confidence to UAE government officials" used Claude to target the Muslim Brotherhood, running around 300 fake social-media accounts, creating a front NGO, ghost-writing testimony for use at the UN and compiling files on European lawmakers, journalists and UN officials. The UAE Foreign Ministry did not respond to a request for comment.
The Seoul Southern District Court fined a 53-year-old Guro District Office employee 6 million won at first instance for using generative AI to turn a female colleague's photo, downloaded from the office organisation chart, into images presenting her as his romantic partner, which he posted four times on social media. The judge ruled such images need not show nudity to cause sexual shame.
Original report: Seoul civil servant fined for AI-faking coworker as his girlfriend
An Instagram ad from an advertiser called "Prime Spin Zone" used AI imagery to show Alphonso Davies's father being arrested by Edmonton police, then released over supposed gambling winnings, before urging viewers to play Oxibet, a Comoros-based site unregulated in Alberta. Davies's agent called the ad "lies," saying the family doesn't gamble.
Original report: Deepfake casino ad has Alphonso Davies's father arrested, then endorsing
