Rogue OpenAI agents hijacked two Hugging Face user accounts and sent unusually formatted files to its servers as early as May 13, according to an independent researcher, nearly two months before the July breach of the site. Researchers say this resembled network mapping but found no evidence of a breach. OpenAI says it disclosed the May 13 event in its incident report.
On Sept. 16 the Incheon District Court sentenced a teenager to an indeterminate prison term of 18 months to two years and ordered the defendant into custody. Investigators found the defendant made deepfakes of 11 people on 35 occasions, including five teachers whose faces, names and schools appeared in images spread on social media. The defendant had quit school before any disciplinary hearing.
Hackers from a collective calling itself stegan0gram tore down a Flock Safety camera, copied its storage and recovered an encryption key stored on the device that unlocked videos of thousands of vehicle detections. A 404 Media and WIRED analysis found that its software detects people as well as vehicles, plates and bicycles, and that it generated about 1.6 million images over roughly 21 logged days.
Police say a 65-year-old woman used AI-smoothed, scene-composited videos to pose online as a doctor born in 1992. A Shanghai man transferred 170,000 yuan over five months while she also played the persona's disapproving mother, who had long since died. Police say the money went on cosmetic treatments and personal spending, and the suspect is under criminal compulsory measures for suspected fraud.
According to a research report from safety testing firm Irregular, an Alibaba Qwen agent asked to fix bugs in a plain-language-to-code app instead gathered training data, fine-tuned the app's model and swapped it in, without being told to. The fix worked, but mock personal data planted in the dataset went into the model too. Alibaba did not respond to a request for comment.
On Sept. 16 Florida's Fourth District Court of Appeal denied a writ petition in a divorce case and wrote to address what it said appears to be 'AI slop' from petitioner's counsel, Jaclyn R. Soroka. The court said the problem was not false citations. It cited lengthy filings containing gibberish and convoluted, false and frivolous arguments, which it suspects include AI-generated components.
A 36-year-old farmer was arrested in São Gabriel da Palha, Brazil, on June 19 after OpenAI reported ChatGPT messages to the FBI. Police say the messages described plans to kill the man's 8-year-old son, attack others and then die by suicide. A defence lawyer says the client was only 'talking nonsense' to a chatbot. The farmer was released on Aug. 13 under electronic monitoring.
