Stress-test an AI promise
Every AI policy says what the system will always do, and what it will never be used for. Type one of those promises below. Six hostile-but-lawful readers will try to break it, live.
What just happened
That is the engine room of the Always/Never Audit, run live on a single line. The real audit points the same council at your whole document: a responsible-AI pledge, a deployment covenant, a use policy, vendor assurance language, the AI clauses in a contract. Where different seats converge on the same gap, you have found a structural problem rather than a matter of taste.
The difference between this and the deliverable: the audit adds a human reviewer with twenty-five years in social-risk audit, who weighs what the council finds against the document and signs a written hardening memo. The council finds; the human judges.