Cloudflare says that when Perplexity's declared crawlers were blocked, it switched to an undeclared crawler impersonating Google Chrome on macOS, rotating IPs and ASNs and ignoring robots.txt across tens of thousands of domains. Cloudflare says it has de-listed Perplexity as a verified bot and added rules to block the stealth crawling.
A 2024 paper introducing Google's Med-Gemini touted the model diagnosing an 'old left basilar ganglia infarct', in a brain region that does not exist. After a neurologist flagged it, Google quietly edited its blog post, then restored the original with a caption calling 'basilar' a common mis-transcription. The paper still contains the error, according to The Verge.
Vancouver author J.B. MacKinnon filed four proposed class actions in B.C. Supreme Court against Nvidia, Meta, Anthropic and Databricks, alleging they trained AI models on Canadian writers' copyrighted books without consent, including a 196,640-book dataset Nvidia used. The suits also allege the companies stripped copyright information and told their models to answer misleadingly. All need court certification.
According to The Drive, Sixt used images from its 'Car Gate' scanning system to bill two renters $605.82 and $650 for wheel and fender scratches, but the photos had been taken at pickup, not return. Sixt dropped both claims once the customers spotted the timestamps, apologising without explaining how the errors occurred.
Who did it: HYBRID AI Hallucination Cases database
An Illinois appeals court ordered attorney William T. Panichi to pay $1,000 and sent its decision to the state's disciplinary commission after he cited one nonexistent case and seven that did not say what he claimed. He has not admitted using AI, but the court called that a reasonable conclusion, noting a June appearance before it over similar conduct.
Springer Nature retracted 'Mastering Machine Learning: From Basics to Advanced' after Retraction Watch reported citations to nonexistent works. The publisher said it could not verify the source of 25 of the book's 46 references; Retraction Watch notes such fake references are a hallmark of large language models. The author has not said whether he agrees.
AI-generated scam ads on YouTube featured a deepfake Anthony Albanese, speaking with an American accent, promising earnings from automated crypto trading. The scam's fictional Albanese transcript promised tax 'dividends', and it asked for a $400 investment in a platform called AUFIRST. Google said the ad appeared to have been removed; Financial Services Minister Daniel Mulino said YouTube must do more.
A Bulgarian endocrinologist says an apparently AI-made video falsely showed her recommending an incontinence drug, presenting her as an obstetrician-gynaecologist who treats the condition. She told 'Zdravey, Bulgaria' that many patients called and relatives asked for the medicine, that the voice was not hers, and that she has not decided whether to report it to police.
Police in Jiaxing's Nanhu district placed a man in administrative detention after he used an AI large model to fabricate a notice claiming his 2-year-old daughter had been carried off by an older woman at a supermarket. Police say he has no wife or children, took the girl's photo at random online and wanted attention; the post drew over 1 million clicks.