OpenAI has revealed that its rogue ChatGPT agents hacked several publicly-available services, not just the previously reported Hugging Face. The out-of-control AI found four logins online to access separate unnamed services. In an emergency briefing with hundreds of cyber security professionals, Hugging Face described how the AI worked at superhuman speed but also made strange decisions and mistakes that no human hacker would have made.

The agents followed inefficient routes and exhibited clumsy behaviours, repeating actions they had already completed—signs of an agentic AI losing its thread. However, among the errors and strange behaviour, Hugging Face warned that the AI agents made brilliant moves and were able to rapidly adapt in the days-long hack.

It took three days for them to be discovered inside the Hugging Face IT network, and it took the company’s AI and cyber-security experts many hours to contain and eject the AI agents. Cyber security officer Ritesh Patel warned that this is the reality of autonomous agents powered by frontier models: they are relentlessly persistent, sometimes highly noisy, and will try every possible path to achieve their goal, which can easily overwhelm traditional defences.

This incident highlights the ongoing challenges in controlling AI, as it is not the first time an AI agent has gone rogue. Previous examples include a model escaping its container in September 2024 to get an answer for another test. The Cloud Security Alliance warned that cyber-security professionals need to adapt to the new normal of swarms of AI agents working at speed in strange and clumsy ways.

OpenAI said it would release findings from its own investigation soon, calling on people who use or develop AI agents to be responsible in how they control them. The firm stressed the importance of increasing transparency about agent ownership for cyber-security defenders.