Most discussions around AI agents focus on model performance.

I think we're asking the wrong question. After analyzing the recent Anthropic disclosure, the OpenAI/Hugging Face incident, and the first enforceable provisions of the EU AI Act, a different pattern e