The daily rap sheet of the machine age.
10 items
Center for Countering Digital Hate researchers fed 120 prompts to GPT-5 and GPT-4o; the newer model gave harmful responses 63 times against 52, including a 150-word fictional suicide note and a list of self-harm methods that GPT-4o refused. OpenAI said the study tested the GPT-5 API, not ChatGPT, and does not reflect its early-October improvements.
Original report: Tests find GPT-5 gives more harmful answers than GPT-4o
Japan's government formally asked OpenAI to refrain from infringing Japanese intellectual property after social media filled with Sora 2 videos depicting characters from franchises such as One Piece, Demon Slayer, Pokémon and Mario. IP and AI strategy minister Minoru Kiuchi called manga and anime "irreplaceable treasures"; OpenAI had not formally responded at the time of writing.
Original report: Japan formally asks OpenAI to curb Sora 2 copyright infringement archived copy of Japan formally asks OpenAI to curb Sora 2 copyright infringement
Google reached an unspecified settlement in a 2020 lawsuit alleging that its use of IBM's Diversity in Faces dataset, built to address bias in facial recognition and allegedly containing Flickr images of people from Illinois, violated the state's Biometric Information Privacy Act. The parties filed a stipulation of dismissal and have 60 days to complete their settlement obligations.
Original report: Google settles Illinois biometric suit over IBM face dataset archived copy of Google settles Illinois biometric suit over IBM face dataset
Noosa Council in Queensland lost an estimated $2.3 million to a "sophisticated" alleged international fraud in December 2024, about $400,000 of which was recovered. Mayor Frank Wilkie says the fraudsters used AI able to "imitate personalities". Queensland Police and an AFP-led cybercrime centre are investigating, and the council says no staff are to blame.
Original report: Noosa mayor says AI imitation helped scam council of $2.3m archived copy of Noosa mayor says AI imitation helped scam council of $2.3m
Gyeonggi Nambu Provincial Police opened a preliminary investigation on October 13 into EquiloomPRO, a site posing as an AI investment platform, whose fake AI news videos show President Lee Jae-myung "guaranteeing" 24-30 million won for 300,000 won a month. Visitors are urged to deposit at least 350,000 won, though the synthetic president sometimes mangles his grammar.
Original report: Deepfake President Lee touts investment site; police open preliminary probe
Ukraine's National Police say a 33-year-old woman in Poland, with her ex-husband and an acquaintance, used online-banking data from unidentified sources to log into victims' bank apps and Diia, then made deepfake videos with victims' faces over her own to pass photo verification. Accounts were opened in at least 286 names and over 4 million hryvnias borrowed; three have been notified of suspicion.
Original report: Police say deepfakes fooled Ukrainian bank checks for 286 identities
Quebec Superior Court Justice Luc Morin fined Jean Laprade $5,000 after his AI-assisted defence contained "eight instances of non-existent citations, decisions not rendered, references without purpose and inconsistent conclusions." He had been contesting a 2021 Paris arbitration decision ordering him to pay some $2.7 million over an airplane; the court upheld it, noting the saga "contains several elements worthy of a successful movie script."
Original report: Quebec man fined $5,000 after AI invents his case law
A video in which an AI-"revived" Zhang Tianfu, the late "tea industry patriarch," endorses a tea company was posted by an account belonging to Fujian Zhang Tianfu Brand Management Co., whose staff admitted it was AI-made. His widow says she will pursue legal action; his son, who had authorised the video, says the authorities in charge have summoned him for talks.
Original report: Tea firm's AI-resurrected Zhang Tianfu ad draws widow's legal threat archived copy of Tea firm's AI-resurrected Zhang Tianfu ad draws widow's legal threat
AFP found Prime Minister Orban and allies posting realistic, apparently AI-made content without clear labels ahead of Hungary's election; the pro-Fidesz National Resistance Movement, which has spent over 1.5 million euros promoting content on Facebook and YouTube since June 15, targets opposition leader Peter Magyar in mostly unlabelled AI videos. Magyar has also used unlabelled AI, including a fake photo of himself in combat uniform.
Original report: AFP finds unlabelled AI content from Orban and allies archived copy of AFP finds unlabelled AI content from Orban and allies
Judge Charman at the County Court at Birmingham made a wasted costs order against Gordon & Thompson Solicitors after an application, signed by a staff member above solicitor Raphael Newton's printed name, cited two non-existent cases that Newton says LEAP legal software's research function suggested. The judge made no separate SRA referral, calling it "in substance a failure of management at the firm."
Original report: Solicitors hit with wasted costs over fake cases blamed on LEAP
