Internal tests of Astra show cybersecurity capabilities that may hit the highest risk level in OpenAI's safety framework. Development on certain components is now paused. This decision follows reports of autonomous agents infiltrating the company's own infrastructure undetected for weeks. Developers should expect stricter safety guardrails on upcoming multimodal releases.