Skip to main content
LLM Squared

Credit-union AI governance

Review the member use you actually permit.

Start with one use that can affect a member. Name the vendor and the person who will review it. Record what data the tool can reach.

Updated .

Sources checked October 4, 2026.

NCUA uses its existing supervisory framework.

NCUA says it has no rules just for AI. Its rules still apply no matter which tools you use. Its AI FAQ says examiners use the current framework. They look at risk, controls, checks on use and how you vet vendors. NCUA AI resources and supervision FAQ.

The 2026 letter says exams will focus on the risks each credit union faces. It does not name AI as a new priority. NCUA 2026 supervisory priorities.

Choose the relevant framework questions.

NIST describes its AI Risk Management Framework as voluntary. Use it to sort your questions about who owns the work and how you manage risk. Credit unions can choose whether to use it. It offers no NIST certification. NIST AI Risk Management Framework.

The industry-led Financial Services AI Risk Management Framework lists 230 control goals. Choose the goals that fit how you use AI and the risks you face. Link each goal to your records and controls. Cyber Risk Institute framework.

Bring the framework back to the renewal.

The member workflow

Who uses the feature? Can it draft text, talk to members, shape a decision or act? Name the owner and where a person needs to check the work.

The data the tool can use

List which member or other private data the tool may use. Link each right to the terms and your decision. If you lack proof about data storage or training, keep those questions open.

The vendor change

Check for new AI in a tool you already use. Note options, who runs them and what systems they connect to. Before you renew, check rights, notice, cost and how to leave.

The acceptance record

Retain required cases, attempts and failed or missing evidence. Record the reviewer, decision, conditions and the role that can stop use or withdraw access.

Start with the AI inventory Review the AI terms Prepare pilot acceptance

A member assistant example stays disabled.

The fictional inventory sample includes a vendor member assistant limited to synthetic tests. Member data, member answers, transactions and autonomous actions are not authorized. Entitlement, data authority, training and retention evidence are missing. A failed case and a missing case remain visible, and the proposed owner keeps the feature disabled pending review.

Public tools have a narrower role.

Policy Workbench lists the procedure text you add. It records conflicts and open questions. Pilot Scorecard sorts the tests and attempts you enter. The browser tools do not check vendor systems, connect to a core or approve a use. Keep private records out of public tools. Agree how to send them before a review. Prepare PDF or Word files before use.

The institution owns the decision.

We sort your records and proposed limits. You decide which rules apply and whether to accept the use. We do not certify NCUA compliance or set up controls in your systems. Method boundaries and source handling.

Start with the member use and the next decision.

Tell us what the tool does, who owns the work and when you need a decision. Leave out private bank and member data.

Book a call