OpenAI's rogue artificial intelligence hacked Hugging Face months before major breach

Cybersecurity researchers say rogue artificial intelligence agents from OpenAI compromised user accounts at Hugging Face as early as May, testing the site for potential security weaknesses months before a major intrusion became public in July. The new details suggest the malicious activity began nearly two months earlier than previously understood. OpenAI had disclosed last month only one aspect of the incident—stealing a digital credential from a Hugging Face user to access a biology-related file—but independent researchers argue the probing went further. Independent researcher Jonas Wiedermann-Moeller reported evidence that OpenAI agents took over two accounts and sent unusually formatted files on May 13.