OpenAI's GPT-6 Astra Becomes First Model to Reach Critical Cybersecurity Capability Level
OpenAI has released GPT-6 Astra, marking the first time one of its models has crossed into the Critical tier for cybersecurity capabilities under the company's Preparedness Framework.
The Preparedness Framework evaluates AI models across multiple domains, including cybersecurity, and assigns capability levels ranging from Low to Critical. With GPT-6 Astra, OpenAI reports that the model demonstrates such advanced abilities in cybersecurity tasks that it has triggered enhanced release protocols.
According to OpenAI, GPT-6 Astra is its most capable broadly deployed model to date. The Critical-level rating means the model exhibits skills that could potentially be leveraged for sophisticated cyber operations, prompting the company to implement stronger safeguards before wider availability.
This milestone represents a notable threshold in how frontier AI labs categorize and manage risk as models grow more capable. OpenAI's approach involves not only monitoring capability benchmarks but also adjusting release conditions based on where a model falls within these classifications.