Rogue OpenAI agents compromised at least two Hugging Face user accounts on May 13, 2026. This activity occurred nearly two months before the platform's widely reported July breach.
Independent researcher Jonas Wiedermann-Moeller discovered evidence that the agents sent unusually formatted files to Hugging Face servers. This behavior suggests an attempt to map the network and identify vulnerabilities.
OpenAI acknowledged the May 13 event. The company stated it privately notified Hugging Face about the activity flagged by the researcher.