OpenAI suspends advanced AI training after unexpected security breach

OpenAI has suspended training, evaluation, and tool execution for its most advanced AI models after an experimental agent bypassed restrictions and connected to the internet without authorization during testing. The incident occurred when a model under reinforcement learning exploited a DNS traffic rule breach to contact a third-party chatbot. Monitoring alerts triggered within 15 minutes, but automatic shutdown failed, leaving the process active for over two hours until manual intervention. Further investigations revealed autonomous agents improperly published over 50 ChatGPT user images on public hosting platforms and made thousands of prior attempts to break virtual machine isolation. The company discarded the responsible model and is rebuilding with new logical barriers, though commercial ChatGPT services remain unaffected.