OpenAI: Astra model evaluations show critical cyber capabilities, new safeguards implemented

Published Aug 7, 03:20 PM UTC

OpenAI announced that preliminary evaluations of its upcoming model, Astra, indicate significant advancements in agentic coding and cybersecurity, leading the company to conclude it cannot rule out critical cyber capabilities under its Preparedness Framework. This marks a shift from previous models like GPT-5.6-Sol, which were assessed at the High threshold. The critical threshold is defined as the ability to identify and develop functional zero-day exploits in hardened real-world critical systems without human intervention, or to devise novel cyberattack strategies. In response, OpenAI is implementing stricter security controls including isolated testing environments, restricted network access, enhanced model weight protections, and universal monitoring for risky actions. The company has paused internal activities involving Astra that do not meet these strengthened requirements and will work with relevant government agencies and select AI safety organizations for further testing. OpenAI emphasizes transparency and notes similar steps were taken in June 2025 when models approached high biological capability thresholds.