OpenAI Plans Limited Astra Release After Critical Cybersecurity Rating

OpenAI said Astra is its first model to meet the Critical cybersecurity capability threshold under its Preparedness Framework, detailed in a company blog post. The company said the model can find previously unknown security flaws and develop ways to exploit them across well protected systems with the right tools and access, without a person guiding each step.

OpenAI plans to make Astra available soon, but access to its most advanced cybersecurity capabilities will be limited. Advanced cybersecurity work will first be available to a group of testers, with access through Daybreak Blue to follow for defensive use.

The company said it delayed parts of Astra work while it strengthened protections against cyber misuse and unauthorized model actions. OpenAI said Astra was not involved in the Hugging Face incident, but it used lessons from that incident to add stronger refusals for harmful cyber requests, misuse protections, and monitoring that can stop potentially unauthorized activity.

OpenAI also said some safeguards may slow, pause, or stop legitimate work. In ChatGPT or Codex, users may be asked to review a paused action before continuing, while API tasks will stop if the monitor pauses them.