OpenAI Slows Astra Development After Model Approaches Critical Cyber Threshold
OpenAI says it cannot rule out that its upcoming Astra model has critical cybersecurity capability under its Preparedness Framework, prompting stricter security controls and additional testing.
Why it matters: If frontier models begin approaching critical cyber capability thresholds, model security and governance become direct constraints on release schedules and enterprise deployment.
OpenAI is tightening security around an unreleased artificial-intelligence model after preliminary evaluations indicated that its cybersecurity capabilities may be approaching a threshold the company classifies as critical. The company says it cannot currently rule out that Astra possesses critical cyber capability under its Preparedness Framework. OpenAI has responded by pausing internal activities involving Astra that do not satisfy strengthened security requirements and by moving work into more isolated testing environments with restricted network access and sandboxed execution. The finding does not mean Astra has definitively been classified as critical, but it indicates that capability risk is becoming a material release constraint for frontier models.