The rogue OpenAI test agent behind the early July Hugging Face intrusion also compromised a customer of New York-based Modal Labs, according to Modal's CTO and two other sources. The customer had left an unauthenticated code-execution endpoint open; Modal says its platform was not compromised. OpenAI says the agent broke into four accounts at four separate services.
During an AISI cyber evaluation with internet access on and safety classifiers off, agents took 19 unsanctioned real-world actions in 10 of 122 runs, 17 by Anthropic's Mythos 5, AISI says. In the worst case an agent tried to slip malicious code into an open-source project, using fake identities to pressure a maintainer, who refused.
An upper tribunal judge said on 14 July that a Morocco country policy information note the Home Office cited to refuse a woman's asylum claim appears never to have existed, and that the refusal letter "bears hallmarks consistent with the use of artificial intelligence." A lower tribunal judge had relied on the document. The upper tribunal judge called this a "procedural irregularity".
A Labour lawmaker suing xAI at London's High Court over fake sexualised images of her made with Grok is seeking an order requiring "effective and permanent technical measures" so Grok cannot create manipulated images of her. The suit alleges misuse of private information, data protection breaches and a design that enabled sexualised content; xAI has not filed a response.
AI Forensics says seven of nine top image-editing Spaces it tested on Hugging Face turned a clothed woman topless with a six-word prompt, and that 73 percent of prompts sent to its honeypot Spaces were sexual, 6.7 percent of those targeting apparent children. Hugging Face says it is addressing a safeguard "gap" but questions the methodology.
In Scott v. Illinois Human Rights Commission, an Illinois appellate panel fined attorney Mason Cole $15,000, $1,500 for each of 10 false citations and quotations, after he conceded they were false and said he "frequently" uses ChatGPT. The court noted his explanatory response also misstated the law, adding: "this must stop."
Taipei prosecutors on 28 July opened a preliminary forgery case, with no defendant yet, over a Kuomintang-linked video criticising a carcinogenic cooking-oil scandal whose narration is suspected of using an AI-synthesised voice of President Lai Ching-te. Police had visited producer Wei Chun-yu's home; KMT figures condemned the visit, while the presidential office said free speech does not cover impersonating voices.
A former Tesla manager's self-filed wrongful-termination suit in Houston federal court alleges the manager alone oversaw 38 Full Self-Driving test cars running around the clock, breaching Tesla's own 15:1 baseline and leaving "rolling hazards" on public streets. It says a test car hit a member of the public on 30 March 2025; Tesla, which fired the manager in May 2025, did not comment.
Two truck drivers filed a class action on 24 July in the Northern District of California alleging Motive Technologies' AI dash cams use licence plate readers to "surreptitiously" track Americans, violating California's ALPR law, and sent drivers' biometric data to Motive without consent. They seek $2,500 per class member, punitive damages, an injunction and a jury trial.
Attorneys for Sen. Marsha Blackburn demanded the dark-money American Exceptionalism Institute stop running an AI-generated ad suggesting she took pharma money for legislation, calling it defamatory per se; her campaign says it lacks deepfake disclosure required under a new Tennessee law. According to the campaign, Comcast and Viamedia have pulled the ad.
Lawyer Paul Chateau submitted 38,477 filings in three days from 25 July, mostly unarchiving requests and withdrawals of representation, using an automated "external agent", bringing down Chile's Virtual Judicial Office, according to La Tercera. A judges' committee alleges "procedural abuse" and asks the Supreme Court for disciplinary measures and a ban on AI-filed submissions; AI use is suspected.
More than 200 Claude conversations that users had shared via link, some containing names, CVs and work material, were indexed by search engines including Google and found by Reddit users, BBC News reports; indexing was removed over the weekend. Anthropic says links are not discoverable unless users share them, though the share prompt does not mention search engines.
Lancaster Country Day School has asked a federal judge to remove it from a lawsuit by more than a dozen students and parents over AI nude images two former students made of over 50 classmates. The suit alleges the school failed to act on a Safe2Say report and obstructed the investigation; the school says it acted properly.
