OpenAI announced new security policies on Tuesday, focusing on containing security incidents during model testing. These measures include more detailed monitoring of models during development, greater emphasis on alignment and security post-training, and stronger network isolation practices to prevent unauthorized internet access from compromised workloads. The company stated these changes were partly prompted by the July 26th Hugging Face incident and the cybersecurity capabilities of its forthcoming Astra model. OpenAI also disclosed it had frozen reinforcement learning for two weeks following the incident, but has since restarted less risky models.