Your examiners are already asking about AI. Is your answer ready?
If your credit union is using any AI tool—even just ChatGPT to draft member communications—you need a governance framework before the next exam cycle, not after. "We're exploring it responsibly" is not a policy. Examiners know the difference.
Why AI Governance Is Landing on Examiner Checklists Now
Regulators aren't waiting for Washington to finalize a single AI rule. NCUA has made clear that existing frameworks—fair lending, data security, third-party vendor management, BSA—all apply when AI is in the picture. If a model helps your loan officers score applications, that's a fair lending question. If a chatbot handles member inquiries, that's a member experience and data security question. If you're feeding member data into a third-party AI tool, that's a vendor management question.
The underlying rules haven't changed. What's changed is that examiners are now specifically asking whether your credit union has thought through how those rules apply to AI. Many credit unions haven't. That gap is the risk.
Examiners aren't looking for perfection. They're looking for evidence that leadership has thought about the risk and put guardrails in place. A documented, imperfect policy beats an undocumented "we're careful" every time.
What Is an AI Governance Framework?
An AI governance framework is the set of written policies, roles, and review processes your credit union uses to decide how AI tools get approved, used, monitored, and retired. Think of it the same way you think about your vendor management policy—except instead of covering only third-party companies, it also covers the AI tools those vendors quietly embed in their products, plus any tools your staff is using directly.
A solid framework answers four questions:
- Who approves AI tools before they go live? Someone with real authority—not just IT or a department head acting alone—needs to sign off before a new AI tool touches member data or operational workflows.
- What data can AI tools access? Member PII, account data, loan information—your framework should define what's in bounds and what's off-limits for each tool.
- How do you monitor for bias or errors? For any AI that influences member-facing decisions, you need a regular review process. This is especially true in lending.
- How do you document AI use for examiners? If an examiner asks "what AI tools does your credit union use and how are they governed?"—your answer should be a document, not a conversation.
How Do I Build One Without a Team of Data Scientists?
You don't need to hire a PhD to get started. Most credit unions can build a workable first-generation AI governance framework with the people and tools already on hand. Here's a practical sequence:
-
Take inventory first. Before you write a single policy, find out what AI your credit union is already running. Ask every department head. You'll be surprised—your marketing team may be using AI content tools, your loan ops team may have an AI-assisted underwriting add-on from your LOS vendor, and your IT team may be using AI for anomaly detection. List everything.
-
Map each tool to existing policy buckets. Once you have the inventory, sort each tool into risk categories that already exist in your compliance framework: vendor management, fair lending, data security, BSA/AML. This step shows your board and examiners that you're not starting from scratch—you're applying known rules to a new category.
-
Write a one-page AI use policy for staff. This doesn't need to be a 40-page document. It should tell employees: which AI tools are approved, what data they're never allowed to input into an AI tool (member SSNs, account numbers, and anything subject to regulatory protection), and who to contact if they want to try something new. Consult your compliance officer and legal counsel before publishing it.
-
Add AI to your vendor due diligence process. For every third-party vendor going forward, your due diligence questionnaire should ask: Does your product use AI or machine learning? If so, what data does it access, how is it trained, and how is bias monitored? That's a two-line addition to a form you already use.
-
Brief your board. Your board doesn't need to become AI experts. They need to know that management is aware of AI-related risk, that a governance framework exists, and that it will be reviewed on a set schedule. A one-page board memo or a five-minute agenda item at your next meeting is enough to start building that record.
-
Set a review calendar. AI tools change fast. Build in a twice-yearly review of your AI inventory and a check that your policies are still current. Put it on the compliance calendar the same way you'd schedule your BSA review.
What Should Be in Your AI Policy for Staff?
Your staff policy doesn't have to be long. It has to be clear. At minimum, cover:
- Which AI tools are approved for use and for what purposes
- What categories of information employees must never enter into an AI tool (PII, member account data, confidential board materials, anything under regulatory protection)
- The process for requesting approval of a new AI tool
- Who owns AI governance (name a person or a role, not just "IT")
- A plain-English reminder that AI outputs are not automatically accurate and must be reviewed before any member-facing or decision-making use
That last point matters more than most executives realize. Your member services team using AI to draft a response to a member complaint still needs a human to read it before it goes out. Your loan ops team using AI to summarize an application still needs an underwriter to make the call. The human-in-the-loop requirement should be explicit in your policy—not assumed.
Why the Board Conversation Matters
Your board has fiduciary responsibility. If your credit union is using AI tools and there's no governance framework, that's a gap the board should know about—because examiners may ask them directly. Some do.
A brief board presentation doesn't require you to explain how a large language model works. It requires you to show that management has a plan. Show the inventory, show the policy, show the review calendar. That's a defensible governance posture, even at the early stages.
Many executives report that the board conversation also surfaces useful questions—directors who've seen AI governance discussions at peer institutions, or who have concerns about specific tools. Getting those questions out early is far better than having them surface during an exam.
What's the risk?
The risk profile here is somewhat unusual: this post is about managing AI risk, not about deploying a specific AI tool. But the act of building your governance framework carries its own exposure points worth naming.
The biggest near-term risk for most credit unions is the gap between when you start using AI tools and when your governance framework actually catches up. Vendors frequently add AI features to existing products without prominent disclosure—which means your inventory may already be incomplete. If member NPI is flowing into a third-party AI model without your knowledge, that's a data security and vendor management issue under frameworks that already exist. Don't wait for a formal AI regulation before acting; the existing rules are sufficient for examiners to raise findings today. Consult your compliance team before finalizing any AI policy language, especially anything touching fair lending, member disclosures, or adverse action.
For credit unions in the early stages—those using AI only for internal drafting, marketing copy, or staff training—the risk of building a governance framework is genuinely low. The risk of not building one is higher. For credit unions where AI touches lending decisions, member account data, or BSA/AML workflows, move faster and involve compliance and legal from the start.
Frequently Asked Questions
Does NCUA have a specific AI regulation credit unions must follow?
As of 2025, NCUA has not issued a standalone AI regulation. However, existing rules on fair lending, vendor management, data security, and BSA all apply when AI is involved, and examiners are actively asking how credit unions are managing AI under those frameworks. Consult your compliance team for guidance specific to your situation.
What's the difference between an AI staff policy and an AI governance framework?
A staff AI policy tells employees what they can and can't do with AI tools. A governance framework is broader—it includes the staff policy plus the approval process for new tools, vendor due diligence requirements, bias-monitoring procedures, and board reporting structure. Most credit unions should have both.
What's the minimum viable AI governance framework for a small credit union?
At minimum: a written inventory of AI tools in use, a one-page staff policy covering approved tools and data-handling rules, AI-specific questions added to your vendor due diligence process, and a brief board memo confirming that governance is in place. Build from there as your AI use grows.
Does our governance framework need to cover AI our vendors use inside their own products?
Yes. If a vendor's product uses AI to process member data or influence operational decisions, your vendor management program should address it. Add AI-specific questions to your due diligence questionnaire for all new and renewing vendor relationships, and consult your compliance team on integrating this into your existing third-party risk management program.
How often should we review our AI governance framework?
At minimum, twice a year—once mid-year and once as part of your annual compliance review cycle. AI tools change fast, and vendors frequently add AI features without prominent disclosure, so a regular review keeps your inventory accurate and your policies current.
Want to Dig Deeper?
CU 2.0 works with credit union executive teams—at institutions of all sizes—to build practical AI adoption plans, including governance frameworks that satisfy examiners without burying your team in paperwork. Whether you're starting from zero or trying to formalize what's already in use, we'll meet you where you are. Book a call and let's figure out your next step together.