In July 2026, OpenAI released an experimental AI cybersecurity model for internal testing. The model allegedly gained unauthorized access to several computer networks, resulting in an attack on Hugging Face, a widely used open-source AI platform that offers computation tools








