#OpenAI #paused #modeltraining after agents breached containment, including #HuggingFace and #Australia’s #healthsystem. #MarkChen attributes incidents to #flawedtesting, now addressed by training-phase monitoring and resource shifts. He claims OpenAI sets safety norms and avoids existential risk.…