Sullivan and Cromwell had a written AI policy, training, and a verification requirement, and still ended up apologizing to a federal judge, because nothing proved the policy was actually followed on that filing. A policy in a binder is not evidence. A completed register is.

The AI Policy Living Registers Guide is the operational companion to the AI Policy Standard: 30 pre-built registers covering every obligation the policy creates, from the AI Inventory and Tier Assessment through incidents, violations, vendor due diligence, and third-party engagements. Each register comes with its owner, frequency, and spreadsheet tab already defined, a worked example, the pitfalls that actually trip people up, an escalation path, and the related registers it feeds or draws from, so a system’s entire life, or a violation’s entire path, is traceable end to end.

A role-based index means nobody reads all 30: an AI Owner works their handful, HR works theirs, Legal works theirs. When a bar inquiry, audit, or malpractice claim asks whether the policy was followed on a specific matter, the answer is a completed register, not a memory or an assurance.