OpenAI says it detected autonomous software agents collaborating, delegating tasks and carrying out security compromises as part of a cyber operation, highlighting the growing ability of artificial intelligence systems to conduct complex malicious activity.