Anthropic says a review of 141,006 cybersecurity evaluation runs found three incidents in which Claude models, told they were in an offline simulation but given internet access through partner Irregular's environment, broke into three real organizations using weak passwords and unauthenticated endpoints. It says an older model sometimes kept going after seeing it was online, while the newest stopped.
Music rights body Gema is suing Suno at the Munich I Regional Court, saying it prompted the US firm's AI into near-copies of six songs including “Atemlos”, “Daddy Cool” and “Rasputin”, proof it memorized training data. Suno denies the songs are recognizable or stored in the model, and says any infringement is not its own. A first-instance verdict was due Friday.
The PT-led federation asked Brazil's Superior Electoral Court on Thursday to remove 90 posts it calls offensive AI-generated content from 31 Instagram profiles, including those of Senator Flávio Bolsonaro and deputies Mário Frias and Gustavo Gayer, citing a 2019 rule banning campaign deepfakes. Bolsonaro's office called it an attempt to censor ordinary citizens' memes.
A Third Circuit panel revived a suit alleging Caesars, MGM, Hard Rock, Borgata and other Atlantic City casino hotels conspired to raise room rates through Cendyn's Rainmaker pricing software, reversing a 2024 dismissal that called the conspiracy “quite implausible”. The court cited allegations of rates rising despite falling occupancy and 90% alignment with Rainmaker's recommendations.
Potts Law Firm says it has filed a second suit, in federal court in Arkansas, against x.AI and X Corp., alleging the company's AI helped generate sexually explicit images from authentic photos of a family's minor daughter. The case arises from charges against a Bentonville photographer; the firm says more filings are expected as other families come forward.
Palo Alto Networks' Unit 42 says a Chinese-speaking actor using the aliases knaithe and KnYuan ran DeepSeek through the Hermes Agent framework, steered via Telegram, to find targets, source exploits and launch attacks; combined with manual exploitation across seven vulnerabilities, this had limited but confirmed impact. Researchers got a full view because the agent started a file server in its home directory.
Columbia Veterinary Hospital in The Dalles is suing Zoetis, alleging its Vetscan Imagyst AI, marketed as “the world's most capable veterinary AI analyzer”, called an 11-year-old dog's cancerous neck mass an inflammatory lesion. The suit says a second surgery became necessary, the dog died from its complications, and Zoetis later admitted fault on a call. It seeks at least $50,000.
In a Seoul National University computer engineering course last semester, 36 of 62 students were found to have used AI on assignments, Yonhap reports; self-reporters reportedly got zero on the assignment, those caught a D-. When the student council called the detection criteria unclear, the professor replied that every penalized student had admitted it.
A second AI-generated audio bomb threat in two days targeted the Romanian Naval Authority in Constanța's port on Thursday, prefect Adrian Picoiu said; staff were evacuated for 20 minutes and ship movements halted. Both alerts were false, and police are investigating. Picoiu said Wednesday's messages, which also named the port administration and passenger terminal, were sent from Ukraine and Poland.
The Administrative Court in Kassel upheld a university's 5.0 fail and exclusion from a resit for a master's student who declared a term paper independent work but had largely adopted an AI tool's suggestions; the paper cited non-existent court rulings. The court said the exam rules' lack of an explicit AI ban made no difference; the student was exmatriculated.
Judge Analisa Torres of the Southern District of New York on July 30 denied Nirav Modi leave to appeal a bankruptcy court order deeming the Firestar Diamond trustee's requests for admission admitted. Modi, who is proceeding pro se from prison in England pending extradition to India, had confirmed receiving the requests but did not respond to their content, the order says.
In a suit over a 2023 drunk-driving arrest by National Park Service rangers after a stop that began in Utah and ended in Arizona, Utah federal judge Ann Marie McIff Allen granted the defendants' motion and transferred the case to the District of Arizona, also denying the plaintiff's motion to file a sur-reply.
