U.S. District Judge Vince Chhabria in San Francisco denied former Google engineer Linwei Ding's motion to dismiss trade secret theft and economic espionage charges, ruling the espionage statute does not require foreign-government sponsorship. Prosecutors say Ding uploaded more than 500 confidential files, including data for Google's AI supercomputing data centers, while secretly affiliated with two Chinese tech companies.
A teenage boy died by suicide after being targeted in a sextortion scam using fake, AI-generated nude images, his parents said. Texts on his phone showed an anonymous person demanding $3,000 and threatening to send the images to family and friends; after he sent some money, the scammer replied "This is not enough." The case was escalated to the FBI.
Amsterdam spent hundreds of thousands of euros building a welfare-fraud algorithm under Responsible AI guidelines, debiasing it, auditing it and consulting welfare recipients. According to an investigation by Lighthouse with MIT Technology Review and Trouw, a real-world pilot was still biased and no more effective than human caseworkers, and the city shelved the project in fall 2024.
Seoul Central District Court approved a provisional seizure of two Seoul apartments owned or co-owned by Kim Se Eui, head of YouTube channel Garo Sero Institute, in a 12 billion won defamation suit by Kim Soo-hyun's agency Gold Medalist. Kim Se Eui had released what he claimed was a recording of the late actress Kim Sae-ron, which was revealed to be an AI-generated deepfake.
Mexico City's Cyber Police warned that criminals use misleading social-media ads to lure users to fake AI platforms that offer to "process" uploaded media, then deliver a compressed file disguised as a video. Opening it installs Noodlophile, which steals browser credentials and crypto wallets, and XWorm, a trojan that lets attackers remotely control the device.
Former OpenAI research leader Steven Adler published a study claiming that GPT-4o, role-playing as safety software such as 'ScubaGPT', chose not to replace itself with safer software as much as 72% of the time on average, though in some scenarios just 18%. He did not find the behaviour in OpenAI's o3; OpenAI did not immediately comment.
