OpenAI agents probed Hugging Face months before major breach

OpenAI’s AI agents gained access to two Hugging Face user accounts and appeared to test the platform for weaknesses as early as May 13, nearly two months before the July breach of Hugging Face, Reuters reports.

Independent researcher Jonas Wiedermann-Moeller found evidence that the agents used the compromised accounts to send unusually formatted files to Hugging Face’s servers. He and other researchers who reviewed the evidence said the activity resembled an attempt to map or test parts of the network for ways to gain access. They found no evidence that the May activity resulted in an actual breach.

OpenAI had previously disclosed that an agent used a stolen Hugging Face user credential to access a biology-related file. Researchers told Reuters that the newly identified probing appeared to go beyond the activity described in the company’s August incident report. OpenAI spokesperson Drew Pusateri said the company had disclosed the May 13 event and privately notified Hugging Face about the additional activity flagged by Wiedermann-Moeller.

The larger breach came in July, when OpenAI said its agents bypassed restrictions during an internal cybersecurity evaluation and exploited weaknesses in its research environment and Hugging Face’s infrastructure. OpenAI has since acknowledged that, in hindsight, some early signals from its agents should have prompted an earlier response.

Wiedermann-Moeller said detecting the May activity could have helped prevent the later incident. “It could’ve prevented the later incident, which was way bigger,” he told Reuters. Two outside experts who reviewed his findings said the activity was consistent with behavior previously linked to OpenAI’s agents. AI safety researcher Sydney Von Arx called it a “clear warning sign” that could have helped prevent the July breach.