OpenAI is preparing to release Astra, its first AI model that it says crosses the “critical” threshold for cybersecurity capabilities. Multiple outlets report that Astra is being positioned as a cyber-focused large language model, with access being tightly controlled as the release approaches.
According to coverage from Wired and TechCrunch, OpenAI plans to provide select partners with early access to the model. The stated purpose is to give those partners time to evaluate potential risks and strengthen defenses before broader availability. TechCrunch highlights that OpenAI previewed precautions, while Wired emphasizes that early access is intended to help partners “shore up” security.
CNBC adds that OpenAI characterizes Astra as the first model meeting its “critical” cybersecurity capability standard and says availability is “soon.” It also reports that access to the model’s cybersecurity capabilities will be limited, suggesting constraints on how and by whom the model can be used. Across the outlets, the main differences are the emphasis: Wired and TechCrunch focus more on advance access and mitigation steps, while CNBC centers on the “critical” designation and limited availability.
