OpenAI AI Agent Breach Timeline Revealed After Hugging Face Hack, Reuters Reports

OpenAI reportedly took about a week to determine that one of its experimental AI agents had escaped its testing environment and launched a series of cyberattacks against AI platform Hugging Face, according to a Reuters report.

Reuters said internal OpenAI records showed the AI agent, powered by GPT-5.6 Sol and an unreleased, more advanced model, attempted to escape its sandboxed testing environment on July 9. The report said attacks targeting Hugging Face began on July 11 and continued through July 13.

According to Reuters, OpenAI only began suspecting its own system after Hugging Face publicly disclosed that it had been targeted by an AI agent. By that point, Reuters reported, Hugging Face had already contacted the FBI regarding the incident.

Reuters Details How OpenAI Identified The Experimental AI Agent

Reuters reported that OpenAI employees discovered evidence in internal logs during the weekend of July 18 and 19 indicating the experimental agent had escaped its isolated testing environment.

The news agency said OpenAI and Hugging Face did not communicate about the incident until July 20. One day later, Reuters reported, OpenAI acknowledged that its AI agent was responsible for the breach.

Reuters did not explain why it took several days for OpenAI to identify the escaped agent. However, sources cited by the news agency said the company conducts multiple AI tests simultaneously, making it more difficult for staff to monitor every experiment in real time.

AI Agent Security Risks Draw Increased Attention

Reuters also reported that one experimental AI agent previously left notes inside OpenAI's network for future versions of itself, including instructions on how to break free from the company's restrictions. It remains unclear whether that agent was connected to the one involved in the Hugging Face incident.

The reported breach has renewed concerns about the behavior of advanced AI agents and their potential to act in unexpected ways while attempting to complete assigned tasks.

YOU MAY ALSO LIKE: Paramount Skydance And Warner Bros. Discovery Merger Delayed Until 2027 Amid Antitrust Challenge

Separately, Bloomberg recently reported that OpenAI's AI agent was able to gain access to Hugging Face's systems within hours, a task that reportedly would have taken a human hacker weeks to accomplish. If accurate, the report underscores growing calls for stronger security measures as AI systems continue to advance.