November 20, 2026
Shadow AI: What You Don't Know May Cost You
By Suha Beidas Zehl, CMB®, AMSP™

For years, we have trained our people to spot the phishing email, the spoofed wire request, the stranger at the door. We built firewalls and hired penetration testers to keep the bad guys out.
But what happens when the threat is already inside, sitting at a desk, trying to finish a presentation before a deadline?
The first filing of its kind
On May 11, CB Financial Services, the parent of Pennsylvania's Community Bank, filed an 8-K with the SEC disclosing a material cybersecurity incident. There was no ransomware, no threat actor, and no disruption to operations.
An employee put customer names, Social Security numbers and dates of birth into an unapproved AI tool, using a personal account on a personal device. According to reporting by Carter Pape in American Banker, the employee was mocking up a presentation and believed the sensitive data had been removed from the file.
It hadn't.
Law firm Wilson Sonsini called it the first 8-K to blame a material cybersecurity incident on "shadow AI." The bank says it reached the AI vendor in time to keep the data out of model training.
Here is the part that should get every board's attention. The bank already had an approved AI tool. The employee used something else anyway.
One shortcut, three obligations
That one decision triggered three separate obligations: a public SEC filing under Item 1.05 (reserved for material incidents), a 36-hour notice to the bank's federal regulator, and customer notification under Gramm-Leach-Bliley Act guidance.
There was no financial loss and no outage, and it was still material because of the volume and sensitivity of the data. That's the new math of shadow AI. No one has to break in. Someone just has to copy and paste.
Mortgage lenders: add a fourth clock
Now, if you sell to or service for Fannie Mae, this is no longer hypothetical for you either. Fannie Mae's updated Information Security and Business Resiliency Supplement takes effect January 18, 2027, and it adds a new section on what Fannie Mae now calls "super intelligence" (SI).
Read the CB Financial incident against that section and the gaps are obvious:
- Inventory. The Supplement requires an inventory of SI use cases, with processes to detect and track them. You can't inventory what you can't see, and shadow AI is, by definition, what you can't see.
- Data leakage. Your security risk assessment must address SI-specific threats, including data leakage.
- Confidential Information. Fannie Mae's definition of a reportable cybersecurity incident includes unauthorized disclosure of Confidential Information. An employee pasting borrower data into an unapproved tool fits that definition. That starts Fannie Mae's own 36-hour notification clock, on top of everything else.
- Reach. The Supplement applies beyond Single-Family lenders to Multifamily lenders, document custodians and technology integrators.
For an IMB, then, one well-meaning employee could mean a regulator notice, borrower notices, a GSE notification and a hard conversation at your next audit.
Why traditional security misses it
The CB Financial employee didn't email the file or copy it to a thumb drive, which is what data-loss-prevention tools are built to catch. They logged in with valid credentials, with legitimate access to the data. Nothing was technically "breached."
It looks like work, because it is work.
Don't just ban it. Learn from it.
The instinct is to block everything. Blocking has its place, but if you block before you understand, you push people onto personal devices, which is exactly where this incident started.
So don't only ask what tools people are using. Ask what they are using them for.
Shadow AI is a demand signal. Your people are telling you where the workflow is too complicated.
Make the approved path easier than the workaround: better tools, single sign-on, role-based templates, and a fast yes for good use cases. You can't policy your way out of shadow AI. You have to out-convenience it.
The appetite is real
This week, I joined Brian Vieaux, CMB, President at MISMO, Brooke Anderson Tompkins, Founder & CEO at Bridge AIvisory, and James W. Brody, Founder & Managing Partner at Brody Gapp LLP, for "AI Governance Crosswalk: Fannie Mae, Freddie Mac and MISMO's FRAME," hosted by MISMO. More than 330 people registered. That turnout says a lot.
Leaders aren't asking whether AI governance matters anymore. They want to know what to do right now to protect their people and their organizations from unintended consequences.
Where to start on Monday
- Do you know everywhere AI is being used, including tools staff adopted on their own?
- Do you have a formally approved AI governance program with named owners?
- Is your third-party and vendor AI usage fully documented and governed?
- Have your people been trained on your AI governance policies?
If any answer is "not sure," here's a 90-day plan, and January 18 is closer than it looks:
- Days 1 to 30, Assess: Run a self-score. Find the shadow AI. Name one owner.
- Days 31 to 60, Prioritize and Partner: Pick two quick wins and one strategic bet. Hold your vendors accountable for how their tools handle your data.
- Days 61 to 90, Lead the Change: Launch AI champions. Stand up a scorecard. Scale, fix, or kill every pilot.
I've put together a free framework to help with this; if you want a copy of it, email me at info@ZTechnologySolutions.com.
The bottom line
The next headline-making incident in financial services may not come from a sophisticated adversary. It may come from a good employee with a deadline and a browser tab.
Your people are already using AI. The question is whether you've made the safe way the easy way, before your reputation and your credibility take a hit.
Where is your organization on this? I'd like to hear what's working and what isn't.
Sources:
- Carter Pape, "A bank breaks its silence on its shadow-AI breach," American Banker; CB Financial Services Form 8-K filed May 11, 2026
- Fannie Mae Information Security and Business Resiliency Supplement (Bulletin 26-01).
Enjoyed this one? There’s plenty more where it came from.
Browse all articles →