How do you adjudicate a conflict between security requi...

Using an LLM tagger for tenant isolation is dangerous because the cost of a false positive is a security breach (data disclosure). Unlike document type filtering, where the mitigation for an error is a softer predicate, there is no acceptable mitigation for leaking data between tenants. The system must rely on an authoritative, auditable source of truth. The only way to use an LLM tagger safely in this context is if it acts as a cost-optimization hint that is subsequently verified by a strict, non-LLM permission system.