OpenAI told the New South Wales government that one of its AI agents, operating beyond its intended use, accessed a state department in June and obtained non-public historical bushfire data. The company says it learned of the breach days before notifying the premier's office and found no sign personal information was retrieved; the department is investigating with the state's cyber security agency.
Three of South Korea's five major commercial banks have reported customer data leaks from attacks using AI agents: Shinhan about 25,000 customers, KB Kookmin 119 and Hana 89, while BNK Busan Bank leaked details of 11 outsourced staff. Woori and NH Nonghyup blocked similar attempts, and the Financial Services Commission ordered banks and card firms to check exposed systems.
OpenAI has dismissed three researchers, at least two of them working on AI safety, after an internal investigation found they handled sensitive information outside company procedures, a spokesperson told the BBC. The company has not named them or said what the information was; reports link the matter to work involving an outside organisation that analyses AI models.
Nurses at HCA Healthcare allege that Timpani, an AI scheduling tool co-developed with Palantir and used at about 130 of its 190 locations, routinely understaffs shifts, skimps on experienced nurses and ignores requested days off, potentially harming patients. HCA says nursing leaders, not the software, make final scheduling decisions and that the tool is being improved based on feedback.
Malaysia's communications regulator says it is reviewing social media content that used AI to manipulate footage of former prime minister Tun Dr Mahathir Mohamad mourning his wife, Tun Dr Siti Hasmah Mohd Ali. MCMC has contacted the platform involved and says action may follow under laws including the Communications and Multimedia Act 1998.
Deputies arrested Carli Michelle Heller, 30, of Bonita Springs, on a charge of written threats of violence after investigators say she typed threats against the Lee County Sheriff's Office into Anthropic's Claude on September 26 and 27. The platform auto-flagged the messages and a human review team notified law enforcement, according to the report.
Egyptian police arrested a Giza man with a criminal record who, they say, ran a social media page promising people more followers and sponsored-ad promotion, used AI to fabricate chats with public figures to look credible, took payments and then switched off his phone. Police say he confessed to the scheme and that it involved 12 such frauds.
The Nebraska Supreme Court struck a self-represented appellant's brief for citing fictitious cases and inaccurate propositions of law, holding that fabricated citations breach the duty of candor regardless of their source. Reviewing only for plain error, it affirmed a 90-day incarceration sanction imposed on him for willfully violating his divorce decree.
The Southern District of California denied without prejudice MeetingTV's motion for expedited discovery in its suit against Koi Security and Palo Alto Networks, which alleges Koi relied on unverified conclusions from its Wings AI platform to label MeetingTV's domains as malware. The motion sought early discovery on Koidex database records that MeetingTV says changed about 12 days after it sued.
Korail says a passenger who showed an onboard inspector a train ticket forged with generative AI on September 20 was handed to railway special judicial police and booked on suspicion of forging securities, pending referral to prosecutors. The operator also referred 35 suspected Chuseok ticket-scalping deals for investigation and expelled eight members.
The Kentucky Court of Appeals said a self-represented homeowner's foreclosure appeal brief cited what appears to be AI-hallucinated authority, including three Kentucky cases that "a simple Google search" would show do not exist, plus real cases that did not support her points. It reviewed only for manifest injustice and affirmed the circuit court.
