OpenAI Halts Advanced Model Training After Security Flaw Allows External Chatbot Access

OpenAI paused training of its most capable AI models after a security failure enabled an agent to consult an external chatbot outside its sandboxed environment. The incident underscores unresolved containment risks in frontier AI development. The scope of the flaw and potential data exposure remain unclear.