On September 1, OpenAI officially labeled its soon-to-launch Astra model with a 'Critical' cyber capabilities rating under its Preparedness Framework—a first for any of its AI systems. According to the company, this rating specifically applies to the Daybreak Blue extended access configuration. After launch, Astra's most advanced security features will only be available to a select group of users.

What Happened

OpenAI assigned Astra a 'Critical' level for cyber capabilities in its Preparedness Framework, marking the first time the developer has given such a status to one of its own AI models. Astra is still in pre-release, with the system currently being prepped for launch.

Astra's Cyber Capabilities

OpenAI says that with the right tools and access, Astra can identify previously unknown vulnerabilities and figure out ways to exploit them in protected systems—all without step-by-step human guidance. This skill set is exactly why the model got the 'Critical' label in terms of cyber risk.

How Access Will Work

The test results apply to the Daybreak Blue extended access configuration. OpenAI says the most advanced security features will remain restricted post-release, only available to certain users. The model is still in the pipeline, and the company emphasizes it will tightly control access to its most sensitive capabilities.