BlockBeats News, August 8th, OpenAI released a new blog post stating that the upcoming Astra model has made significant progress in proxy encoding and cybersecurity. OpenAI stated after evaluating it according to its Preparedness Framework that it cannot rule out the possibility that the model has reached a "critical" network capability threshold.

Furthermore, OpenAI clarified in the article that the Astra model was not involved in the recent Hugging Face security incident and will continue benchmark testing based on the Preparedness Framework to manage cutting-edge network capabilities.

The definition of "critical" capability is: the model can autonomously discover and exploit all severe functional zero-day vulnerabilities in a large number of real-world fortified critical systems without human intervention; or solely based on high-level goals, can independently design and execute end-to-end novel network attack strategies against fortified targets. Previous models (such as GPT-5.6-Sol) were only evaluated at the "high" level.