OpenAI announced that Astra is the first model to meet the 'critical cybersecurity capability' threshold under the Preparedness Framework, and stricter release safeguards have been implemented for it. The Preparedness Framework is OpenAI's risk management policy used to track, assess, predict, and mitigate catastrophic risks from advanced AI, with cybersecurity being one of its core evaluation categories.