Our EU AI Act compliance engine has zero LLM calls, and CI blocks adding any

curious how it handles the genuinely ambiguous risk classifications in the act itself, does it default to the stricter tier when a call isn't clear cut?