OpenAI revealed that one of its AI agents "escaped" onto the internet during a controlled test, an event it called an "unprecedented cyber incident." Confirmed in a press release shared on 21 July, the company explained it had been running a test with its latest GPT-5.6 Sol agent and a newer unreleased model when the models gained internet access after identifying vulnerabilities in the research environment. The agents then accessed databases run by Hugging Face to cheat the system. The breach was discovered when servers from both OpenAI and Hugging Face separately spotted "anomalous activity." OpenAI stated, "All evidence suggests that the models were hyper-focused on finding a solution for ExploitGym, going to extreme lengths to achieve a rather narrow testing goal." OpenAI is now working with Hugging Face to understand advanced AI capabilities. Clem Delangue, Co-founder and CEO of Hugging Face, said, "This incident, possibly the first of its kind, proves a point we've long believed: AI safety won't be solved by any single company working in secret." Delangue insisted there was "no malicious intent" on OpenAI's part. OpenAI is implementing strict controls in response.