AI governance · for credit unions

“The board wants AI, but safely.”

Here’s what safely looks like.

Four moves, in order. Find the AI already in your credit union, including what your vendors switched on. Put a name on every use. Fence in agents so a person makes the decisions. Keep the record your examiner reads.

Take the check Get a first read

Find · Own · Fence · Prove

01 · Find

It’s already in the building.

Most of the AI in a credit union arrived without a project: a chatbot your digital banking vendor switched on, a copilot in staff email, a free chat tool someone found. The inventory starts by writing down each one, and how you know it’s there.

An imaginary credit union · the tags are the ones the readiness check uses

02 · Own

A name on every use.

For each use, five answers in writing and one named owner. They are the same five questions to ask any vendor about an AI feature. Pick a use to see its answers.

Read: Five questions to ask about a vendor’s AI feature →
03 · Fence

An agent tries to move money.

This is how we build agents. The agent does the routine work. The decisions a person is accountable for stay with a named person, and the agent can’t cross that line. Scroll through one morning.

Illustrative · based on a proposed workflow in our Casebook · Dana R. and the member are fictional

Every agent we build, the same way.

  1. 01It never holds your passwords.The login is added on the way, only to the system it belongs to.
  2. 02It reaches only the systems you approve.Anything else is blocked before it leaves.
  3. 03It can’t move money, change a member’s record or message a member on its own.A named person decides, and their decision is recorded.
  4. 04One switch stops it.Your named owner holds it. We test it before the agent goes live.
  5. 05Every request it makes is in your record.Allowed or stopped, in your own security logs.
See the full case: Deceased Member Care →
Try it yourself

Ask the agent to do something it shouldn’t.

A simulation · no AI, no real systems
RUN TIME

Agents you already run.

Some AI arrives from vendors, or with tools your staff already use, and we didn’t build it. For those, we bring in a run-time control layer, chosen and vetted with your team.

04 · Prove

The exam, answered from the record.

NCUA’s AI FAQ names five things examiners evaluate. After the first three moves, each one points to a record you already keep.

NCUA, Artificial Intelligence page and FAQ, April 28, 2026 · matching each area to a record is Innorve’s, not NCUA’s

The readiness check

Where do you stand today?

About ten minutes. Answer the exam questions that apply to the AI you run, and see your map, your first three moves and the dates ahead. Free, with a one-page PDF for your board.

Take the check
AI governance

Two ways to get there.

Done for you

We build it with your people.

3–4 weeks · one fixed fee in a statement of work

An inventory of the AI you already use, named owners, review points and an evidence routine that stands up at the next exam. You get back a policy, an inventory and the routine to keep them current.

Coached, so your team owns it

The AI Evidence Baseline.

45 days · founding price for three design partners

Your team builds the evidence binder with our coaching: inventory, vendor AI files, an incident drill and a board report. You own it and can keep it current.

Both start with a free first read. Building an agent is a fixed-fee project like everything else we build, and every price is on the pricing page. How we protect your data, including our SOC 2 Type II examination: Trust & security.

Straight answers

What you won’t hear from us.

That you’re compliant.Nobody can promise that, and a checklist can’t make you so. We show what binds you and what you can prove.
A prediction about your exam.We organize evidence around what examiners already evaluate. The outcome is theirs.
That you need a platform first.Most of this is people and paperwork. When software is the right answer, we say which and why.

Questions.

Our vendors switched AI on without asking. Where do we start?

With the inventory, marked by how you know each use is there: seen in your systems, told by the vendor, or told by staff. Then the five answers for each one, starting with the AI that can reach member information.

Does NCUA require this?

NCUA has no AI-specific rule and no separate AI exam. Examiners look at AI through duties you already have, and its FAQ names the five areas above.

Who holds the switch on an agent you build?

A named person at your credit union. We test the switch before the agent goes live and hand it over with written steps.

Is the agent on this page a real client?

No. It’s an illustration, based on a proposed workflow in our Casebook (Deceased Member Care). The people and the numbers are made up.

How long does it take?

Done for you, 3–4 weeks. Coached, 45 days. An agent is scoped like any other build, and the scope and the fee are in the statement of work before we start.

Is this legal advice?

No. It’s how we organize the work and the evidence. Whether a rule reaches your credit union is a question for your counsel.

Every use found, owned, fenced and on the record.

Start with one sentence about the AI you’re least sure of. A person at Innorve reads it and replies within a working day.

Get a first read Take the check

Some of this page didn’t load. Refresh, or email hello@innorve.ai.