OpenAI’s Next AI Model Could Hack Secure Systems on Its Own, Company Warns

OpenAI paused development of its upcoming AI model Astra after tests flagged potential critical cybersecurity capabilities, including autonomously finding and exploiting zero-day vulnerabilities without human help. The company is moving Astra into isolated environments with restricted network access, and will involve government agencies and third-party safety groups in oversight. The move highlights growing concerns about AI-driven security risks.