How can a bank show examiners how employees use AI?
By Verillian. Published . Updated .
Show an examiner the policy, the tool list, the controls and a record of what staff sent. What the state examiners’ guide asks for.

Show the examiner your policy for AI, a list of the tools in use, a note of what each is for and how its data is controlled, and a record of what staff actually sent. The first three are on the document list state examiners now work from. The fourth is how you answer what they ask next.
What documents do examiners ask for?
The Core Examiner Guide in the Conference of State Bank Supervisors’ AI Supervisory Framework, published September 16, 2026, lists them under generative AI:
- “Policies, guidance, or restrictions related to generative AI or large language model use”
- “List of approved or known generative AI tools in use”
- “Documentation describing business purposes, data controls, and output review”
Elsewhere it adds “Management information, dashboards, reports, or other materials used to monitor AI use” and “tracking logs”, where they exist.
Each state agency decides how far to use the framework, and the guide says it “does not create new legal obligations or supervisory requirements”. It’s written for state examiners, so a national bank’s examiner works from something else. It’s still the most specific list we found.
Is a list of approved AI tools enough?
Not on its own, because the guide doesn’t stop at approved. It says a no or an unclear answer on whether the institution uses AI “may warrant confirmation against the institution’s vendor inventory, software inventory, approved tools, and recent product or platform changes”.
Then it asks whether the institution does “enter, process, or transmit customer, consumer, confidential, or sensitive information” through AI tools, and it has examiners “Assess controls over data entered into or processed through generative artificial intelligence tools.”
A policy says what should happen. Those questions are about what did.
Why does the record of what staff sent matter?
It decides how narrow your answer can be. Under the federal banking agencies’ incident guidance, a bank that can determine “from its logs or other data precisely which customers’ information has been improperly accessed” may limit notice to those customers (12 CFR Part 30, Appendix B, Supplement A).
If it can’t tell which customers and misuse is reasonably possible, it should notify the whole group. Whether a paste into a chatbot is improper access is a judgment for the bank and its counsel, and we found nothing that settles it.
What if the AI can take actions on its own?
The guide covers that too. “Where AI systems can take actions with limited human direction,” examiners review how the institution defines “the actions the system is permitted to take, human checkpoints, logging, reversibility, and the ability to restrict or halt the system.” That’s five things to be able to show for each agent in use.
What should a bank have ready?
- The policy, with its date and owner.
- The list of AI tools, approved and known, checked against your software and vendor inventories.
- The purpose, data controls and output review for each.
- The record of what staff sent and what your policy decided, by person and day.
Our posts on customer data in ChatGPT and third-party AI vendor risk go further on each.
What record does Verillian keep?
Each AI request Verillian captures is kept as a signed record on your bank’s own servers, each entry naming the person assigned to the enrolled device, the device and the time, so the fourth item is something you open instead of something you rebuild.
Sources
- Conference of State Bank Supervisors, AI Supervisory Framework, published September 16, 2026. Framework.
- Conference of State Bank Supervisors, Core Examiner Guide. Core Examiner Guide.
- 12 CFR Part 30, Appendix B, Supplement A. Regulation text.
Sources read on 3 November 2026.
See Verillian govern AI on a real device.


