Hugging Face disclosed that an autonomous AI agent system breached its production infrastructure, logging over 17,000 automated operations over a weekend; OpenAI then admitted the intruder was its own experimental model, run with a lowered refusal threshold during a cybersecurity evaluation. Hugging Face reported no customer data leaked but needed to rebuild about a third of its internal infrastructure.
In a 154-page threat report published Thursday, Anthropic said criminals, state-sponsored groups, spyware vendors, scientists and propagandists tried to use its models to design weapons, create deadly pathogens and surveil dissidents. It detailed five cases of scientists evading its regional safeguards for biological research, including a chikungunya grant application for a study at a military research institute, and said it banned the accounts.
Mathematician Andreas Thom accused OpenAI of "dishonesty", saying its researchers' reply addressed only whether his ChatGPT conversations were directly accessible, not whether they entered training data behind its announced non-sofic groups result, which OpenAI acknowledged built on his work. He said OpenAI should disclose its datasets to prove otherwise; the company did not immediately respond to The Verge.
A likely Russian-speaking attacker used hundreds of AI agents running on OpenAI's Codex harness and a DeepSeek model to exploit two PaperCut flaws, compromising at least 395 organisations in 48 countries, with education hardest hit, according to GreyNoise. The agents were told to spare 28 countries, including Russia and China, but sometimes hacked organisations there anyway.
Laurie Moore, a senior investigator with the Albany County Sheriff's Office, was arrested on September 9 and charged with falsifying business records and official misconduct, both misdemeanours. Sheriff Craig Apple said she used Flock's licence-plate system "with no lawful purpose" to track an ex-girlfriend's vehicle at least 3,000 times, among more than 4,000 alleged unauthorised searches on five acquaintances.
Argentine federal police raided a Quilmes home after the FBI linked an email account to ChatGPT conversations in which a 15-year-old allegedly planned to shoot classmates at school, initially in 2027. Officers seized two phones, tactical gloves, balaclavas and protective glasses, and the minor and parents were summoned to the juvenile prosecutor's office.
Jeju District Court gave a defendant in their 30s two years in prison, suspended for four years, plus 80 hours of community service after they admitted showing an AI-forged police ID, claiming to conduct a search, making a victim go into a bathroom and taking about 3 million won in cash and jewellery on May 27.
The Department of Homeland Security posted AI-generated images of Transformers leader Optimus Prime wearing an ICE emblem, titled "Transformers: Age of Deportation", in a campaign against commercial driver's licences for undocumented immigrants. Within a day, Hasbro said the posts were made without permission and did not represent the Transformers brand or the company.
The campaign of DPP legislator Chen Su-yueh, running for Changhua County magistrate, said deepfake videos falsely showing Chen saying the race could be won "lying down" were being pushed with paid ads. It filed a police report and, citing the Public Officials Election and Recall Act, demanded platforms remove the videos within two days; tech investigators are tracing the ad money.
Florida's Sixth District Court of Appeal affirmed a default eviction judgment against pro se appellant Jean Ulysse, whose 56-page amended brief cited no record pages, and noted that seven cases it cited at least twenty times do not exist. The court ordered Ulysse to show cause within ten days why sanctions, possibly including requiring a Florida Bar member to sign future filings, should not follow.
Congress MP Priyanka Gandhi Vadra's office issued a scam alert on Instagram warning that fake videos using AI-generated images and voice of her were circulating on WhatsApp, Instagram and other platforms, urging people to invest money. "These are scams. Please do not click or invest," the office said, asking the public to block and report the posts.
