Bank and credit union marketers adopted AI faster than any other department, mostly on accounts IT never set up. One community bank just learned what that costs, and boards are starting to ask.
Ask most bank or credit union executives where AI lives in their institution and they’ll point to the fraud model, the chatbot pilot, or the vendor demo the CIO keeps forwarding. They usually won’t point at marketing. They should. Marketing teams write with AI, design with AI, and buy media from platforms that rewrite ads with AI, and a lot of that happens on a team credit card or a personal login. In most institutions, marketing is running the largest AI program in the building, and it’s the one with the least oversight.
That gap is about to matter. Boards are adopting AI themselves and only beginning to govern it, and the first SEC filing to blame a material cybersecurity incident on an employee’s unapproved AI use came from a community bank employee building a presentation. When your board or your compliance officer asks for a list of every AI tool your department uses, what data goes into it, and who can turn it off, you want to hand it over the same day. Most marketing leaders couldn’t today.
Bank Marketers Adopted AI on Their Own Accounts
The ABA’s annual survey of bank marketers shows how fast this moved. Reported AI use in marketing went from 16.9% in 2024 to 29.9% in 2025 and 50.4% in 2026. The tools are the ones you’d expect: Microsoft Copilot and ChatGPT tied at 69%, followed by Canva at 58%. It’s a small sample, 116 respondents and all banks, but the direction matches every other source we’ve seen.
The more important number is how marketers get those tools. Only 38% use enterprise platforms their institution provides. Another 47% rely on individual or team subscriptions, and 15% of AI users have no employer-provided generative AI tool at all.
That means most AI use in bank marketing runs through accounts IT didn’t provision and probably can’t see.
Translate that into your own department. A team subscription on a manager’s corporate card has no single sign-on, no data retention settings your security team reviewed, and no audit log anyone checks. Personal accounts are worse: the prompts, uploads, and outputs live in a consumer account that walks out the door when the employee does. If someone on your team has pasted a member segment, a campaign results file, or a draft rate sheet into one of those tools, your institution has no record of it and no way to get it back.
One Employee, One Presentation, One SEC Filing
This stopped being hypothetical in May. On May 5, 2026, Community Bank, the subsidiary of Pennsylvania-based CB Financial Services, discovered that an employee had handled non-public customer information with an unauthorized AI application. The exposed data included customer names, Social Security numbers, and dates of birth. On May 7 the company declared the incident material, and it filed a Form 8-K on May 11. The law firm Wilson Sonsini called it the first 8-K to blame a material cybersecurity incident on shadow AI.
What the employee was doing should stop every marketing leader cold. CEO John Montgomery told American Banker the employee was mocking up a presentation. The bank already offered an approved AI tool with bank-issued accounts. The employee used an unapproved tool on a personal account instead, uploading the file from a personal device while believing the sensitive data had already been removed.
There was no hacker, no outage, and no expected material financial loss. American Banker reported that the single mistake still likely set off three separate obligations: the public SEC filing, a notice to the bank’s federal regulator under the 36-hour rule, and notification of affected customers under Gramm-Leach-Bliley guidance.
Now think about how often your team builds a presentation from customer or member data. Campaign results for the board. A segment analysis behind the CD special. A cross-sell list for the branch managers’ meeting. Each starts as an export with names and account details in it, and each is exactly the kind of file someone hands to a chatbot to clean up on a Thursday afternoon before a Friday meeting. Community Bank did the obvious thing right by offering an approved tool, and it still ended up filing with the SEC. A policy on paper doesn’t stop someone working against a deadline. Access, data controls, and a culture where people ask first do.
Your Vendors Turned AI On Without Asking You
Even a marketing team that never opens ChatGPT is using AI, because the platforms you buy from have built it in. Most of those features arrived as product updates, not as contract amendments, and many are switched on by default.
Meta is the clearest example. Its Advantage+ creative enhancements can automatically change ad copy, images, video, layouts, and calls to action, and in many ad accounts some of those enhancements come preselected. The ad a member sees may not match the creative your compliance officer signed off on. Don’t assume a financial-services ad account is exempt. Look at what’s actually toggled on in your own.
The compliance math is simple: an ad the platform rewrote is still your ad. If an enhancement trims the APY disclosure off a CD promo or rephrases a rate claim, the Reg DD and UDAAP exposure belongs to your institution, not to Meta. Your compliance review approved a file. What ran may be different, and nobody on your team may know.
The same goes for the rest of the stack. Your CRM probably added predictive scoring. Your marketing automation platform probably writes subject lines. Your social scheduler drafts captions, and your website chatbot may have moved from scripted answers to generated ones. Each of those raises the same three questions: Was it turned on without your sign-off? Is your customer or member data training the vendor’s model? Can you turn the AI off without losing the product?
Credit unions should take this one especially seriously. The GAO reported in 2025 that NCUA lacks the authority to examine technology service providers even as credit unions lean on them for AI-driven services. GAO first recommended that Congress grant it in 2015, and as of the report Congress had not acted. America’s Credit Unions opposed the recommendation. Whatever your view of that fight, the practical result is that no federal examiner is reviewing your martech vendor’s AI practices on your behalf. Your contract and your own due diligence are the only checks. Banks have more regulatory backstop, since the FDIC, Fed, and OCC can examine certain service providers, but a regulator’s reach doesn’t change the data-use terms you signed.
Boards Are Behind Too, Which Is Marketing’s Opening
Institutions have moved past experimenting. Cornerstone Advisors found that 59% of credit unions and 49% of banks have already deployed generative AI, and 17% of credit unions have deployed agentic AI. Oversight hasn’t kept up. In Bank Director’s 2026 governance survey, 60% of respondents said their board had adjusted AI-related policies, reporting, or oversight, while 21% said AI risks aren’t overseen by the board at all.
Directors are also using the tools before governing them. Board software vendor OnBoard surveyed 531 governance professionals across industries and found director AI use jumped from 69% to 92% in a year, while 63% of boards still have no formal AI policy and only 6% have one that’s enforced. That’s a vendor survey and it isn’t specific to banking, so treat it as directional. It lines up with the banking-specific data.
We think marketing leaders should read this as an opportunity, not a threat. Boards are about to ask these questions and most don’t yet have a framework for the answers. The department that arrives with a clean inventory, named owners, and a one-page risk summary sets the template the rest of the institution follows. The department that arrives empty-handed gets a policy written for it by someone who can’t tell a Canva resize from a generated background, and that policy will almost certainly be a ban your team quietly ignores.
Not All Marketing AI Carries the Same Risk
Governing this well means sorting it, because treating every AI use the same either buries your team in approvals or waves through the dangerous stuff.
At the low end is internal drafting: brainstorming headlines, summarizing a webinar, outlining a blog post, with no customer or member data involved. That needs an approved tool and a basic acceptable-use rule, and not much more.
In the middle is external content a human reviews before it ships: ad copy, social posts, email drafts, images. The AI is a writer, and your existing compliance review is the control, as long as what ships is what compliance saw. That’s exactly where default-on platform enhancements break the chain.
At the high end is anything that touches member or customer data or acts without a person in the loop: uploading lists or exports, personalization engines choosing who sees a loan offer, chatbots answering rate questions, and automated ad tools deciding audiences and creative on their own. These touch fair lending, UDAAP, Reg DD, and privacy law directly. Each one needs a named owner, a written description of what the system is allowed to do, and a way to shut it off fast.
That last part gets overlooked. If your chatbot starts quoting last month’s CD rate, who can turn it off, how quickly, and can you reconstruct which members saw the wrong number? If the answer is “open a ticket with the vendor,” you don’t have a control.
There’s a newer wrinkle on the horizon. As members and customers start sending their own AI agents to compare rates and move money, the accuracy of your rate pages and product data stops being a content issue and becomes a control issue. We covered the deposit side of that in AI Agents Are Coming for Your Idle Deposits.
What to Do Before the Board Asks
1. Build the marketing AI inventory this month. Have your marketing ops lead or digital manager list every AI tool and feature in use, with five columns: tool, who uses it, how it’s paid for (enterprise, team card, or personal), what data goes in, and whether output reaches members or customers. Include AI features inside software you already license. This is the document your board will eventually ask for, and building it is how you find the exposure you don’t know about yet.
2. Get everyone off personal accounts. If 15% of AI users in the ABA survey lack an employer-provided tool, assume some of your team does too. Buy enterprise seats for whatever tool they’re actually using, or move them to the approved one, and set a firm date after which personal accounts for work are prohibited. Take the cost to your CFO next to the CB Financial filing. A few dozen seats cost far less than an 8-K, a regulator notice, and a round of customer notification letters.
3. Audit default-on AI in every ad platform. Your paid media manager or agency should open each Meta ad account this week, review which Advantage+ creative enhancements are active, and run the same check on every other platform you buy. Turn off anything that can change copy or disclosures on rate, APY, or loan ads. Run the decision past compliance and ask them to check one thing in particular: whether any approved creative ran in a modified form, and what changed if it did.
4. Write a one-page data rule a coordinator can follow. No customer or member names, account numbers, Social Security numbers, or dates of birth in any AI prompt or upload, ever. Exports used for presentations get de-identified or aggregated before they leave the core or CRM. Name the one person who approves exceptions. The Community Bank employee thought the file had been scrubbed, so require a second set of eyes before any customer-derived file goes near an AI tool.
5. Give every high-risk AI use an owner and an off switch. For your chatbot, personalization engine, or AI-generated content program, write down who owns it, who can turn it off, and what triggers a shutdown: a wrong rate, a complaint pattern, a missing disclosure. Then test the shutdown once. If nobody can turn it off within an hour, that’s your first finding.
6. Send three questions to every martech vendor. Have procurement or your vendor manager ask in writing: Have you added AI features to our service, and were they enabled by default? Is our customer or member data used to train your models? Can we disable the AI features without losing the core service? For credit unions in particular, the vendor’s written answers may be the only oversight record that exists.
7. Put one AI slide in the board deck. Every quarter, report the number of AI tools in use, the share of users on enterprise accounts, any incidents or compliance exceptions, and tools added or retired. That turns marketing from the department the board worries about into the one that answered first. If you’re already working on tying marketing reporting to outcomes the board cares about, The Measurement Gap covers how to frame it.
How You’ll Know It Worked
Track four numbers. The share of your team’s AI users on enterprise accounts should reach 100% within a quarter. The inventory should be complete and reviewed by compliance, and no tool should turn up later that wasn’t on it. Every ad account should have a documented enhancement audit, repeated whenever the platform ships a major update. And compliance exceptions tied to AI-drafted or AI-modified content should be counted every quarter and trending down. If you can put those four numbers in front of your board, you’ve answered its AI question before anyone had to ask.
Related Marketing Banks Coverage
- AI Agents Are Coming for Your Idle Deposits
- The Measurement Gap: Bank Marketing Heading Into 2027
- Google Analytics Will Now Drop Traffic From Domains You Don’t Approve
