9
No one is coming to audit your AI because they are curious. They are coming because your firm now runs on systems that process confidential data, and the people who answer for your firm — regulators, insurers, clients, auditors — need to know that those systems are safe. The question is not whether they will ask. It is whether you can answer.
Most firms cannot. Not because the work is bad, but because the answers were never written down. The AI usage exists. The documentation does not. And in an audit, undocumented is indistinguishable from unmanaged.
The questions they will ask
Put yourself in the seat of the examiner. The first question is the simplest and the most revealing: What AI does your firm use, and where does it run? If the answer is a list of browser tabs and a vague sense that people use tools, you have already failed the audit. The question demands an inventory — every tool, every vendor, every deployment, and the location of every piece of infrastructure that touches firm or client data.
The second question follows immediately: What data goes into it? This is where privilege, confidentiality, and regulatory exposure converge. The examiner wants to know whether client material, personal information, or protected data ever leaves your control. If you cannot trace the data flow, you cannot answer — and you cannot defend.
Then comes the question of control: Who can access the system, and what can they do? Access controls, role separation, and audit trails are not optional features. They are the evidence that your AI is governed like every other system in your firm, rather than operating as a shadow process no one supervises.
And finally, the question that separates the prepared from the exposed: How do you know? Not what you believe, but what you can prove. Logs, retention policies, deletion procedures, incident response plans. The examiner does not want your assurances. They want your records.
Why the answers matter more than the technology
Notice what the examiner is not asking. They are not asking which model is newest or which vendor is most popular. They are asking about governance — because governance is what survives scrutiny. A regulator does not need to understand the architecture of your AI. They need to understand that you do, and that you have built the controls to prove it.
This is the audit-readiness gap that most firms have not yet confronted. The technology moved faster than the governance. The tools were adopted in the ordinary course of getting work done, and the policies, inventories, and controls were never built to keep pace. The result is a firm that is using AI everywhere and can account for it nowhere.
Closing the gap
The fix is not to stop using AI. The fix is to make your usage provable. That means building the inventory, mapping the data flows, establishing the controls, and documenting the answers before the question is asked. It means running your AI where you can actually account for it — inside your own walls, on your own hardware, under your own control — so that the answer to every question is something you can demonstrate rather than something you hope.
When the examiner asks what AI you use, where it runs, what data goes in, and how you know, the only good answer is one you can prove. The firms that will pass that test are the ones that treated the question as inevitable and prepared accordingly. The firms that will not are the ones still hoping the question never comes.
Start with the Confidential AI Exposure Checklist at 9tailsadvisory.com/checklist/. Nine lives. Zero leaks.