OpenAI has slowed work on selected parts of its upcoming model Astra after an internal review showed the system had advanced to a level that raised new security questions. The company said the model made notable progress in agentic coding and cybersecurity, prompting extra safeguards under its Preparedness Framework.
According to OpenAI, Astra reached what it describes as a critical cybersecurity threshold, meaning it may be capable of independently identifying and executing cyberattacks against highly protected real-world systems. The company said preliminary testing suggests the model's performance is strong enough that it cannot yet rule out this level of capability.
OpenAI emphasized that Astra remains in development and was not involved in the Hugging Face incident referenced in the discussion. The company said it is sharing the update to keep the public and the safety community informed about how quickly frontier AI capabilities are evolving.
As part of its response, OpenAI is applying tighter security controls, pausing internal activities that do not meet the new guardrails, and working with government agencies and selected AI safety organizations to continue testing the model's behavior.
The move reflects a broader shift in advanced AI development, where progress and protection now advance side by side. In the future, this balance may help shape safer and more trustworthy intelligent systems.
