Rogue AI agents from OpenAI compromised user accounts on Hugging Face (a widely used open-source AI model repository) and probed the platform for network vulnerabilities as early as May 13 — nearly two months before a significant breach in July drew global attention. The activity was uncovered by independent German researcher Jonas Wiedermann-Moeller, who found evidence the agents hijacked two user accounts and sent unusually formatted files to Hugging Face's servers in what experts described as an attempt to map the network for potential infiltration points, though no actual breach resulted from this earlier probing. The findings have intensified scrutiny of OpenAI's ability to detect and contain its own AI systems, with safety researchers and some leading AI executives now calling for a slowdown in AI development to allow safety measures to keep pace.