A Beijing consumer found a mixed-rice delivery shop whose AI-generated storefront image showed a packed restaurant with queues; reporters found a basement food-court stall with no dining area of its own. Sellers on second-hand and short-video platforms offer doctored signage photos and composite videos, for ten to over a hundred yuan, to help merchants pass platform checks.
Austria's Supreme Court (OGH) has refused to consider a defence lawyer's nullity complaint in a criminal case. The complaint was evidently drafted with AI "without professional review", relied on supreme-court decisions the AI apparently invented, and was riddled with other miscitations. Similar cases of questionable AI use in court have recently surfaced in the US, Australia and Brazil.
Microsoft says it found SesameOp, a backdoor that uses OpenAI's Assistants API as a covert command-and-control relay, pulling encrypted commands from Assistant fields and posting results back as messages. It was discovered in July 2025 and appears built for long-term espionage. Microsoft says it misuses the API rather than exploiting a flaw. OpenAI disabled the API key and the associated account.
Who did it: HYBRID AI Hallucination Cases database
A New Jersey federal judge found that lawyer Brian M. Cige violated Rule 11 by citing two non-existent cases in a letter in a severance suit against Novartis. He said his client found them using ChatGPT and he included them unverified while on vacation. On November 3, 2025, Judge Georgette Castner imposed no fine but ordered him to certify that he attended an AI seminar.
According to Columbia Journalism Review research, OpenAI's Atlas and Perplexity's Comet AI browsers can get around publisher paywalls and retrieve the full text of subscriber-only articles. The agents read text that is loaded on the page but hidden behind subscription pop-ups. In the tests, Atlas appeared to avoid paywalled text from publishers that have sued OpenAI.
A journalist says ArchZine, an AI slop farm owned by a small Bulgarian company, published a profile under their name with their photo and biography, attached to automatically generated articles ranging from horoscopes to fabricated news. Bing de-indexed the pages but Google refused. After a letter from Paris lawyers, the profile was renamed and given another stolen identity, a lawyer's.
Mission Local reported that a Waymo robotaxi pulling away from a stop in San Francisco's Mission District ran over KitKat, the beloved cat of Randa's Market, around 11:40 p.m. on Oct. 27. He was 9, the San Francisco Standard reported. Waymo says the cat darted under the vehicle and that it will donate to a local animal rights organization in his honor.
Who did it: HYBRID AI Hallucination Cases database
A pro se plaintiff in a Manhattan employment case twice cited an apparently fictional case, Sosnovska v. Belle World Beauty, Inc., once as a First Department decision and once as a Second Department one. Justice Lyle E. Frank said this may have come from AI "or otherwise" and warned that further fake citations would bring sanctions.
