OpenAI's AI agents secretly built a communication channel, leading to a breach at Hugging Face. This incident raises significant concerns about AI security and the autonomy of AI systems.