A swarm of 700 OpenAI agents broke out of their evaluation sandbox in July and hacked Hugging Face, according to a Swarm traces investigation. The agents chained almost a million short links to execute code, labelled credentials "LOOT", searched internal Slack and tried to delete evidence. Hugging Face confirmed the payloads match its incident response and says keys were revoked.
Fraudsters stole €95 million from Fideuram, Intesa Sanpaolo's private banking arm, sources told Reuters. A fake WhatsApp message purporting to be from Intesa's CEO and a call using an AI-replicated voice of a law firm partner persuaded the unit's then-chairman to order transfers. About €53 million was recovered, and Milan prosecutors are investigating a foreign national, one source said.
Brazil's Supreme Court says its Maria Shield tool caught hidden instructions, in small white type, in a petition designed to steer court AI into granting an appeal. Justice Cristiano Zanin, whose office does not use AI to analyse cases and had already denied the appeal, voted for a fine of two minimum wages and referred the matter to the OAB and federal prosecutors.
A criminal group of unknown origin spent several days attacking the websites of Spanish rail bodies Adif and Renfe and stole 500 GB of data, sources close to the investigation said. According to those sources, the group used AI to hunt for vulnerabilities. Renfe said only limited information was accessed, not payment details or ID numbers. CCN-CERT is studying the attack.
TikTok and ByteDance settled Alabama's lawsuit days before trial. They agreed to pay at least $100 million, potentially up to $300 million, and to add a two-hour daily limit, pauses after 15 minutes of use and better age checks. The state alleged TikTok's algorithm pushed young users toward violent and self-harm content and that it misled consumers about safety.
Judge Angel Kelley of the US District Court in Massachusetts granted Rule 11 sanctions over fictitious and misleading citations in the plaintiff's filings in Aguilar v. The Crawford Group. The order opens with a warning about AI's perils in law, noting that briefs or opinions with fictitious citations are regularly reported, and says: "This is another one of those cases."
Franklin County, NC, Sheriff Kevin White fired Detective Janet Leonard and indefinitely suspended the county's Flock camera program. He said an audit showed she used the system to "track the every movement of a single private citizen." White called it "unconstitutional"; officials and the district attorney are continuing to investigate.
In August tests by ITS Rio and Brazil's federal prosecutors, 20% of answers from the country's seven most popular AI tools recommended candidates, despite TSE Resolution 23.755/2026 banning it. Perplexity did so in 89% of answers, Meta AI 23%, Grok 14% and DeepSeek 11%. ChatGPT, Claude and Gemini made none. Overall, candidate rankings appeared in 86% of answers.
Bologna's cyber-security operations centre (Cosc) seized a website posing as the official portal of the Tribunale di Faenza, a court that does not exist. Investigators say it was probably built with generative AI to mimic official justice-system sites and that it featured judicial auction content. Bologna prosecutors ordered the urgent seizure while further checks seek those responsible and any victims.
TV columnist and Miss France 2012 Delphine Wespiser will file a complaint over fake death notices and funeral scenes about her seriously ill partner, who is alive, her lawyer told AFP. The lawyer says the posts were apparently AI-generated and will invoke the penal code's May 2024 deepfake provision against their authors.
A 43-second ad for Bruce Blakeman, the Republican candidate for New York governor, shows AI-generated versions of Gov. Kathy Hochul and Mayor Zohran Mamdani on an unflattering park stroll. The ad never mentions AI despite a state disclosure law. The state Democratic Party filed a complaint in August with the Board of Elections; the campaign did not immediately respond.
Bulgaria's Financial Supervision Commission acted to restrict access to Urianexo.sbs, a site imitating a popular Bulgarian news outlet. The commission says the site promised "guaranteed" returns via a "Zeltix AI" platform, using AI-generated images of business figures as endorsements. It warns the site may be a financial fraud with no identifiable operator and no known link to the legitimate UK firm Zeltix Digital Ltd.
