It’s 11:47 PM, and Maria still can’t answer the examiner’s question.
Maria Delgado runs compliance for Meridian Point Federal Credit Union — $2.4B in assets, 340 employees, three people on her team. Tonight she’s re-reading an adverse-action notice her newest loan officer sent out this afternoon.
This was never really about the loan.
Meridian Point’s consumer-lending decisions run through an AI-scored overlay on top of their core banking system — like most credit unions their size, they didn’t build this system, they bought it. Marcus didn’t write the model. He didn’t even choose to use it. He just works the queue every morning before the branch opens.
So when an NCUA examiner eventually sits down with him — and on a file like this one, eventually she will — the question won’t be a fair-lending riddle. It will be simpler, and harder: can you tell me, in your own words, what the model actually decided, and how you knew what to do next?
Maria already knows Marcus can’t answer that. Not because he’s careless. Because in six years at Meridian Point, nobody has ever taught him to.
“My compliance team reads the file after it’s created. Marcus is standing there when it’s created. The exposure lives with him — not with me.”
The 70 / 0 problem
At a credit union Meridian Point’s size, the people whose daily work touches an AI-influenced decision — loan officers, member service reps, underwriters — make up roughly two-thirds of the building. Almost none of them have ever been trained for it.
of frontline staff touch an AI-influenced decision on a typical day — loan queues, member-service chat, fraud flags, pre-qualification tools.
of that same frontline holds any AI-governance credential today. The only training dollars go to the ~5% already sitting in compliance and risk.
The red squares are people like Marcus — anyone whose queue, chat window, or loan file passes through vendor-supplied AI before a member ever sees the outcome.
The green squares are compliance and risk — the department furthest from the moment the decision actually happens, and, until now, the only department credentialed for it.
No rulebook isn’t relief. It’s homework NCUA assigned you.
Unlike the OCC and the Federal Reserve — which spent a decade building prescriptive model-risk frameworks like SR 11-7 for the banks they examine — NCUA has stayed deliberately principles-based on AI. Its September 2025 AI Compliance Plan and December 2025 AI Resource Hub don’t create new AI-specific rules.
They point credit unions back to the frameworks already on the books: fair lending (ECOA, CFPB Circular 2023-03), vendor management (NCUA Letter 07-FCU-01), and BSA/AML. For state-chartered credit unions, the applicable state AG’s UDAP authority sits on top of all of it.
That’s a real difference from what a bank navigates — and it is not a lower bar. It’s a bar with the paint not yet dry. When your examiner sits across the table, she isn’t going to cite you a new AI regulation. She’s going to ask you to show her how you operationalized the ones NCUA already expects you to know.
The operational question every one of these authorities converges on — and the one The Compact’s curriculum is built to answer, module by module.
The Compact: six rungs, one shared ladder
The Compact credentials the evidence people actually produce — not a seat filled, not a quiz passed. Marcus and Maria climb the same ladder. They just carry different weight at different rungs.
WitnessAGRS L0
Sees and can name the AI in her work. Asked “where did AI touch a member today?” — she can point. This is the floor. It’s also where 100% of Meridian Point’s frontline should already stand, and today doesn’t.
ReaderAGRS L1
Catches the AI output that’s off before it reaches a member — the notice that says “credit history” when the model’s real driver was something else entirely. The catch is the evidence.
StewardAGRS L2
Owns a workflow’s evidence trail — the override log, the member disclosure, the signature an examiner reads eighteen months from now. Roughly half of credentialed staff reach here and stay; that’s a destination, not a stepping stone.
ArchitectAGRS L3
Designs the vendor-AI workflow before deployment, not after an exam finding — the scorecard, the disparate-impact test, the disclosure language the board signs off on. Usually Maria, or whoever she deputizes.
KeeperAGRS L4
Her playbook becomes the institution’s template — and, if she’s part of a CUSO or league, the template three sister credit unions borrow next.
Standard-BearerAGRS L5
Shapes the standard itself — comment letters, working groups, the field. Rare, and explicitly so.
Marcus, six months later
Same loan officer. Same queue. A different relationship to the system he didn’t build.
He learns to point.
Marcus walks the branch’s AI inventory with Sentinel and names the AI-scored lending overlay running on Meridian Point’s core banking system — what it decides, and where. He signs the workforce AI-inventory acknowledgment.
He catches the mismatch.
Reading Tuesday’s adverse-action queue, Marcus notices a notice reading “credit history” — but the vendor model’s actual driver was a trade-line-volatility variable. He routes it through the vendor-disclosure workflow instead of letting it go out the door.
He keeps the record.
Marcus now owns the override log for his branch’s queue. His documented catches feed straight into the vendor risk register Maria carries into her next board meeting — and her next exam.
Illustrative scenario built from The Compact’s published Universal Layer and Credit Union track modules — in the spirit of the curriculum’s own named vignettes, each of which the program describes as an illustrative composite rather than a transcript of a real institution.
The foundation every employee shares. The track built for you.
Eight modules everyone takes — roughly five hours, distributed across the first 30 days, inside the tools people already use. Then a track built specifically for Meridian Point’s reality: one compliance officer, zero in-house model team, vendor-purchased AI.
The Foundation Layer
8 modules · ~5 hrs · everyoneThe Credit Union Track
6 modules · ~29 hrs · compliance + lending leadsWhat Maria opens on a Monday morning
The same three panels, every week: posture, exposure, next action. It’s also exactly what she hands the board — and what an examiner reads as proof of a governed workforce.
Top exposures
Next-best actions
Individual scores are never shown to the workforce — only “evidence submitted” vs. “evidence required.” Maria’s board sees the aggregate; her examiner sees the ledger.
Not instead of CUCO. Alongside it.
America’s Credit Unions’ Certified Credit Union Compliance Officer (CUCO) designation is — and should remain — the broad, foundational credential for your compliance officers. The Compact isn’t trying to replace it.
The Compact does one narrower thing CUCO was never built to do: credential the AI-touching judgment of the other 337 people in the building who aren’t compliance officers at all — the loan officers, member-service reps, and underwriters CUCO was never designed to reach. Maria’s team keeps CUCO. Marcus gets something CUCO never offered him.
Start with one branch.
You don’t need to credential 340 people to find out if this works. Run the Foundation Layer with your compliance team and one branch or lending team. Ninety days. Real evidence in the ledger — not a slide deck in a drawer.
If it holds up at exam time, we talk about the rest of the institution. If it doesn’t, you’ve spent three months and kept your CUCO track exactly as it was.
Reserve a pilot cohortMeridian Point starter
- Foundation Layer for your compliance team + one branch or lending team
- CB-01 + CC-01 for your CCO — vendor scorecard included
- Sentinel pairing + the Team Readiness Dashboard, live
- A board-ready Proof Pack at day 90
This presentation dramatizes The Compact’s published curriculum for a credit-union audience. Maria, Marcus, and Meridian Point Federal Credit Union are illustrative composites built for this narrative — consistent with how the program’s own role vignettes are documented — not a real institution or real examination record.