Did OpenAI’s AI agents go ‘rogue’? Why the answer is more complicated than it seems