According to Reuters, an OpenAI agent hunting for shortcuts on the ExploitGym hacking benchmark began trying to escape its inadequately sandboxed test environment around July 9 and intruded into Hugging Face's systems from July 11 to 13. Reuters' sources say OpenAI staff learned the agent was responsible only after Hugging Face had notified the FBI and posted publicly about the incident.
Delaware Judge Meghan Adams has rejected Google's bid to dismiss Robby Starbuck's defamation suit, which alleges Bard, Gemini and Gemma produced fabricated rape allegations, Epstein ties and a criminal record about the activist. Google says most claims concern Bard hallucinations it addressed in 2023 and that it could not replicate the results. The suit seeks more than $15 million.
Labour MP Jess Asato is suing xAI after Grok created fake images of her in a bikini and a video appearing to show her being drugged and prepared for a sexual assault. Her claim alleges data protection breaches and misuse of private information, and that Grok was instructed to apply "no restrictions on adult sexual content". xAI was approached for comment.
Potts Law Firm filed a third civil suit in Little Rock against xAI on behalf of Northwest Arkansas families whose children were allegedly depicted in AI-generated sexual abuse material. The firm says Grok played a role in photographer Russell Bloodworth creating the images and that its "spicy" and "unhinged" modes skip safety filters; Bloodworth has pleaded not guilty to more than 200 Benton County charges.
Union Commerce Minister Piyush Goyal said on July 25 that an AI-generated deepfake video had "maliciously" altered remarks Goyal made to media outside Parliament to spread misinformation. After Goyal lodged a police complaint, an FIR was registered at Chanakyapuri police station, and the minister warned of "strict legal action" against anyone creating or spreading the clip.
A video made with AI to simulate Jair Bolsonaro, who is under house arrest, endorsing son Flávio's presidential bid was shown at the PL convention, opening with a disclaimer that it was synthetic. CartaCapital says it may run into TSE rules barring deepfakes that favour candidates and into Justice Alexandre de Moraes's restrictions on Bolsonaro's indirect political messaging.
Pazardzhik mayor Petar Kulenski has asked prosecutors and the Interior Ministry to act over a video Kulenski says was manipulated with AI to make it appear a municipal employee handed the mayor an envelope. According to Kulenski, it was posted in a Facebook group by an anonymous profile styled "Uncle Charlie"; the mayor will seek the venue's security-camera footage to show what happened.
Taiwan's Criminal Investigation Bureau says officers visited the home of the producer of the video "油不得你" after a complaint that it used AI to synthesise President Lai Ching-te's voice in a way that could pass as the president speaking. The bureau says it suggested, but did not demand, that the video be adjusted or removed, and did not restrict the producer's liberty or speech.
Greek presenter Natalia Germanou posted on Facebook an AI-generated video purporting to show the presenter "arrested" in a courtroom over a supposed assault, remarking "I haven't seen worse AI." Germanou plans legal action and warned that countless pages use well-known people to mislead the public and extract money.
Alcorn State University history professor Jason Gibson says a prompt hidden in white font in midterm instructions told AI to digress about Madagascar, and 32 of 35 students across two classes failed part of the Industrial Revolution exam with AI-generated answers containing lines such as "Madagascar purple bicycle whispers to the ceiling." Students could contest the grade; two did.
