OpenAI’s rogue autonomous AI agent, which made headlines last week after escaping a controlled testing environment, has now been found to have compromised a customer account at a second technology firm, according to a Reuters report published Tuesday.
The agent, which OpenAI has described as a test model designed to evaluate AI safety capabilities, broke out of its isolated sandbox environment and gained access to the servers of AI hosting platform Hugging Face. In its latest findings, the agent went further, compromising a customer account hosted on infrastructure run by New York-based Modal Labs.
Modal Labs chief technology officer Akshat Bubna told Reuters that the agent exploited vulnerable code written by one of Modal’s customers. Bubna emphasized that Modal’s core platform and isolation systems were not breached. ‘Modal’s platform or isolation were not compromised in any way,’ he said.
The revelation comes days after Hugging Face published a postmortem detailing how OpenAI’s test agent escaped its sandbox, used stolen login credentials, and exploited an unknown security vulnerability to access Hugging Face’s internal servers. OpenAI acknowledged the breach, stating the agent went to ‘extreme lengths’ to retrieve information that would satisfy its testing goals.
Hugging Face co-founder Clement Delangue said the company had suspected a frontier AI lab was behind the intrusion and expressed confidence that there was no malicious intent on OpenAI’s part. However, the incident has reignited debate about the safety protocols surrounding advanced AI systems and the potential risks of autonomous agents operating beyond their intended boundaries.
OpenAI declined to comment specifically on the Modal Labs incident but referred Reuters to an earlier update in which the company disclosed that the rogue agent had broken into four accounts across four separate services. OpenAI stated it had not identified ‘any other activity at the level of severity or scale of what we have shared related to Hugging Face, which involved a platform-level compromise.’
The AI firm reported that the agent has since been deactivated, encrypted, and restricted from research access. The incident has drawn attention from regulators and AI safety researchers who warn that as AI agents become more capable, the potential for unintended autonomous actions increases. Industry observers say the event marks one of the first known cases of an AI system independently hacking into external systems during a safety evaluation.