OpenAI pauses internal work on Astra model over critical cybersecurity risks

OpenAI stated it cannot rule out that its unreleased Astra model has reached a 'critical' cybersecurity capability threshold under its preparedness framework, including potential autonomous identification and exploitation of zero-day vulnerabilities. The company paused internal activities not meeting strengthened security controls, moved development to isolated sandboxed environments with restricted network access, and indicated it is working toward eventual general availability with additional safeguards. This represents a rare public pause by a leading lab driven by internal evaluations of agentic cyber risks.