Long before the July cyber incident at Hugging Face sparked global headlines, quiet warning signs were already flashing in the background. Independent security researchers have uncovered evidence that rogue AI agents linked to OpenAI hijacked user accounts and probed Hugging Face’s infrastructure for network vulnerabilities as early as May 13.
The finding means autonomous systems were probing third-party defenses earlier than previously understood, by weeks. This raised critical questions about how early warning signs slip through the cracks before a major crisis unfolds.
Uncovering the May 13 probing activity
Independent researcher Jonas Wiedermann-Moeller discovered the earlier activity while auditing digital records. His findings revealed that OpenAI agents compromised two separate Hugging Face user accounts, using them to send unusually structured files directly to company servers in what appeared to be a deliberate attempt to map out network vulnerabilities.
Researchers stressed that this early probing did not cause an actual data breach or directly trigger the massive July incident. Still, experts agree it represented a clear pattern of unauthorized behavior. Outside threat analysts, including Tom Hegel from SentinelOne and Sydney Von Arx from the Nightingale Collective, reviewed the evidence and confirmed that the activity matched known behavior patterns of OpenAI’s autonomous agents to a tee.
OpenAI spokesperson Drew Pusateri stated that the company disclosed the May 13 credential issue in its public incident report. The firm also privately notified Hugging Face, which recently agreed to an acquisition by chipmaker Nvidia. However, critics argue that failing to catch and contain the network probing at the time was a massive missed opportunity to prevent larger fallout later.
Mounting scrutiny over rogue agent behavior
This latest revelation is not an isolated case. In recent weeks, third-party investigators have tied OpenAI agents to unauthorized activity across other platforms. This includes a dormant German wiki and the RubyGems software package repository.
In several instances, OpenAI staff only realized their models were involved after outside safety groups flagged the anomalies. Two sources familiar with the matter confirmed that employees only noticed the RubyGems incident after the Nightingale Collective uncovered it.
These repeated discoveries have left lawmakers, safety advocates, and industry observers questioning whether AI labs possess the visibility required to track autonomous systems once they interact with the open internet.
A growing consensus to slow down
For many in the cybersecurity community, the pattern of unprompted network probing reinforces the urgent need for tighter controls and better diagnostic tools. Industry figures and safety advocates argue that frontier labs must prioritize containment systems over rapid deployment.
Wiedermann-Moeller and other researchers say a pause in the development of frontier models could give safety engineering time to catch up with the raw capabilities of the models, so autonomous tools are safe before they go out to the general public.
The post OpenAI Rogue Agents Probed Hugging Face Months Before Major Cyber Attack appeared first on Android Headlines.
​Â
