OpenAI Pauses Astra After It Nears First-Ever "Critical" Cyber Risk

OpenAI said Friday it has paused internal work on its Astra model after the system became the first to reach a potential "Critical" cybersecurity rating under the company's Preparedness Framework. The disclosure came in a company post first reported by Axios. OpenAI's framework reserves the Critical rating for a model that can independently identify and develop functional zero-day exploits against many hardened real-world critical systems or devise and execute end-to-end novel strategies for cyberattacks against hardened targets given only a high-level goal. Every earlier frontier model, including GPT-5.6-Sol, stayed in the High category. Astra's internal evaluations showed gains in autonomous coding and cybersecurity large enough that OpenAI says it can no longer rule the Critical rating out, though full benchmarking is still running. Astra now operates under isolated testing environments, restricted network and tool access, encrypted model weights, sandboxed execution, and chain-of-thought monitoring that can interrupt high-risk activity in real time. Government agencies and selected safety organizations get evaluation access before the public does. OpenAI still plans a broad release once it satisfies the necessary safety and security requirements.