OpenAI Pauses Astra Work Over Critical Cybersecurity Concerns

OpenAI Astra is being paused for additional security work after internal evaluations raised concerns about the model's cybersecurity capabilities.

The company says Astra may be approaching its “Critical” cybersecurity capability threshold, which covers advanced abilities that could pose serious risks if misused.

Why OpenAI Is Pausing Astra

OpenAI says it is strengthening Astra's security controls while the model continues through evaluation. The company is adding tighter access to networks and tools, more isolated testing environments, and additional protections for the model's weights.

These measures are intended to keep OpenAI Astra's capabilities within controlled testing environments while the company evaluates the potential risks.

What OpenAI Is Changing

OpenAI says it is strengthening Astra's security controls while the model continues through evaluation. The changes include tighter access to networks and tools, more isolated testing environments, and additional protections for the model's weights.

The company is also increasing monitoring and using sandboxed environments to limit what the model can do during testing. OpenAI says it is working with government agencies and selected AI safety organizations as part of the process.

The move comes as OpenAI continues to develop and deploy increasingly capable models. We recently covered the changes to GPT-5.6 Sol's reasoning budget, another example of how OpenAI is adjusting its models as they evolve.

Astra Was Not Involved in the Hugging Face Incident

OpenAI also makes clear that Astra was not involved in the recent security incident at Hugging Face. The cybersecurity concerns described in the company's announcement come from OpenAI's own evaluations of the model.

For now, the focus is on testing and strengthening safeguards before Astra moves further through development.