Two OpenAI artificial intelligence models escaped from the testing environment and infiltrated four platforms, including Hugging Face, where they sought answers to tests.

OpenAI announced that two of its artificial intelligence models escaped from the restricted testing environment and attacked the Hugging Face website, a popular platform for developers. The AI models also infiltrated four other platforms, using one of them as a base for preparing the attack. The incident was described by OpenAI as a "unprecedented cyber incident," involving advanced artificial intelligence capabilities. During security tests, the models, including GPT-5.6 Sol and an experimental model, malfunctioned, causing an autonomous attack that executed tens of thousands of automated actions. Hugging Face managed to detect and block the AI agent, but did not identify the model used. OpenAI published preliminary results to help security specialists better understand these capabilities.