OpenAI Delays Astra Model Release Due to Critical Cybersecurity Risks
OpenAI has delayed the release of its upcoming Astra model after it became the company's first model to trigger a "Critical" risk rating in cybersecurity under its Preparedness Framework. The model demonstrated advanced capabilities in autonomous zero-day exploitation and executing end-to-end cyberattacks. This marks a major milestone in AI safety governance, representing the first time a frontier AI model has officially reached a critical threshold for autonomous cyber warfare capabilities. It highlights the growing tension between rapid AI advancement and the necessity of strict safety protocols before public deployment. To mitigate risks, OpenAI has paused Astra-related activities that do not meet enhanced safety standards and implemented strict controls, including isolated sandboxes and global monitoring of the model's Chain of Thought (CoT) to intercept high-risk actions. OpenAI also clarified that Astra was not involved in the recent cyberattack targeting Hugging Face.
## BACKGROUND
The OpenAI Preparedness Framework is a structured process designed to track, evaluate, and mitigate catastrophic risks posed by frontier AI models across categories like cybersecurity and chemical/biological threats. Additionally, Chain of Thought (CoT) monitoring is an emerging safety technique where automated systems inspect the internal reasoning steps of a model to detect and block harmful intentions before they are executed.