A weather report and a forecast are different documents. One tells you what fell yesterday. The other guesses at tomorrow.
Most writing about AI in lending is forecast. This piece is a report. It collects what the public record shows as of late 2026 and sets it beside what 90 lending institutions told us over nineteen months of conversations. Sources are listed at the end.
Adoption moved from pilots to production
Cornerstone Advisors surveyed 416 senior executives at banks and credit unions for its 2026 What's Going On in Banking study. Forty-nine percent of banks said they had deployed generative AI, up from 16% a year earlier. Credit unions moved from 36% to 59%. The same study reports agentic AI reaching board agendas.
Credit is a harder room to enter. In McKinsey's survey of senior credit risk executives, 52% called generative AI a priority for the credit business, while 12% of North American respondents had deployed any use case. Forty-one percent said model validation was holding them back.
Size shapes the path
A September 2026 Economic Letter from the Federal Reserve Bank of San Francisco notes that small and medium-sized banks face greater hurdles in adopting AI than large banks. The authors add a caution for relationship lenders: soft information is harder for current AI tools to process, so adoption could tilt lending away from small businesses.
For a community lender, what the banker knows about the borrower is the franchise. That points to a division of labor. The tool handles the file. The banker keeps the judgment about the borrower.
What the regulators have said
On April 17, 2026, the OCC, the Federal Reserve and the FDIC issued revised guidance on model risk management, replacing the framework in place since 2011. The guidance describes generative AI and agentic AI models as "novel and rapidly evolving" and places them outside its scope. The agencies said they plan a separate request for information on AI.
In practice, a bank that deploys these tools governs them under its own risk management, and examiners can still ask how. Credit unions sit in a similar position. A May 2025 GAO report found that the NCUA's model risk guidance is limited in scope and detail, and that the agency lacks authority to examine the technology providers credit unions rely on.
What lenders told us
Our own study covers 90 lending institutions. It is Texas-weighted, CRE-weighted and built from warm introductions, so it describes the institutions we spoke with and is not a representative sample. You can read the full study here.
Twenty-five institutions described underwriting that lives in Excel and Word. Three put a number on the keying: 60% to 90% of analyst time goes to data entry. Thirteen named memo writing as the bottleneck, and eighteen named post-close monitoring as a blind spot.
The questions they asked most were about trust: 21 asked about data security, 20 asked why they should not use a general-purpose model, and 18 asked about accuracy.
Where the record and the conversations agree
Three points show up in both.
Interest is high and validation is the constraint. McKinsey's 41% and the accuracy and security questions we heard describe the same hurdle from two sides.
The rules for generative AI in credit are still being written, so the burden of proof sits with the lender. That favors tools that tie every figure to a source page and keep a record of who approved each step.
Smaller institutions carry the heavier lift. Work that removes keying and leaves judgment with the banker is the natural place to start.
Three questions for any vendor
Where does our data go, and is it ever used to train a model? Can every figure be traced to its document and page? Who approves each step, and is that recorded?
LenderBox is built to answer those three. If you want to test them on a file of your own, bring a deal.
Common questions
How many banks use generative AI in 2026?
In Cornerstone Advisors' 2026 survey of 416 bank and credit union executives, 49% of banks and 59% of credit unions said they had deployed generative AI.
Does bank model risk guidance cover generative AI?
The revised interagency guidance issued April 17, 2026 places generative AI and agentic AI models outside its scope. Banks govern those tools under their own risk management practices, and the agencies plan a separate request for information on AI.
What do lenders ask first about AI?
Among the 90 institutions in our study, the most common questions were about data security (21), why not use a general-purpose model (20) and accuracy (18).
Sources
- Cornerstone Advisors, What's Going On in Banking 2026.
- McKinsey & Company, Banking on gen AI in the credit business.
- Federal Reserve Bank of San Francisco, How AI Adoption Might Affect Bank Lending, Economic Letter 2026-25, September 21, 2026.
- Office of the Comptroller of the Currency, OCC Bulletin 2026-13, revised interagency guidance on model risk management, April 17, 2026.
- U.S. Government Accountability Office, Artificial Intelligence: Use and Oversight in Financial Services, GAO-25-107197, May 2025.
- LenderBox, What 90 Lenders Told Us About AI, September 2026.



