Persistent prompt injection risks are challenging ChatGPT Atlas, exposing critical security vulnerabilities in agentic AI systems and enterprise deployments.