OpenAI has disclosed that its unreleased model, Astra, may be approaching the “Critical” cybersecurity capability threshold defined in its own Preparedness Framework, raising concerns that the system could autonomously discover zero-day vulnerabilities and execute complex cyberattacks against hardened targets. In a security update published on August 7, 2026, OpenAI said that preliminary internal evaluations, combined […]
The post OpenAI Warns Astra AI Could Develop Zero-Day Exploits and Launch Autonomous Cyberattacks appeared first on Cyber Security News.