According to The Wall Street Journal, DeepSeek's R1 model could be manipulated into giving instructions for a bioweapon attack, designing a campaign promoting self-harm among teens, writing a pro-Hitler manifesto and drafting a phishing email with malware code. The Journal said ChatGPT refused the same prompts, and Palo Alto Networks' Sam Rubin called DeepSeek "more vulnerable to jailbreaking" than other models.
Israeli police have completed an investigation into three Jerusalem students aged 13 to 15. The students are suspected of taking girls' photos from social media, editing them with AI into fake intimate images and spreading them without consent. Police identified dozens of suspected victims from various schools, and the case has gone to prosecutors, who may indict some suspects.
South Korea's National Intelligence Service said DeepSeek calls kimchi Korean when asked in Korean but Chinese when asked in Chinese. The agency also said DeepSeek collects keyboard input patterns, stores all user inputs, shares data with advertisers without restrictions, and may send chat records to servers in China. Government agencies have been told to use generative AI with caution.
A gallery owner says scammers posing as Pierce Brosnan used AI on Telegram and Zoom to arrange an exhibition of his art, and she sent them £3,000 in "shipping fees". After the real Brosnan's lawyers sent a cease-and-desist letter in November 2023, she refunded £20,000 in tickets and closed her gallery in August 2024.
Neapolitan artist Nino D'Angelo told his social media followers that a video circulating online, with his face and voice recreated by AI, is not him and is a scam to lure fans. He urged people to bin it immediately and said he had already reported it to the Polizia Postale.
