← All articles
Regulatory & Compliance

How to evaluate AI lending software under model risk guidance

Vijay Mehra
November 17, 2026

Disclosure. LenderBox publishes this guide and sells AI lending software, so we are one of the vendors it is written to test. It ranks no products. Every regulatory reference below was read on the regulator's own website on November 17, 2026 and is listed under Sources. This guide is not legal or regulatory advice; your counsel and compliance team decide how any guidance applies to your institution. Last reviewed November 17, 2026.

The short answer

To evaluate AI lending software under model risk guidance, start with the guidance that is current: in April 2026 the Federal Reserve, OCC and FDIC replaced SR 11-7 with revised interagency guidance on model risk management. Then decide whether the tool, or any part of it, is a model under that guidance's definition. Where it is, ask the vendor for what the guidance expects of vendor products: enough understanding to judge conceptual soundness, validation, outcomes analysis and ongoing monitoring. Where it is not, the third-party risk guidance still applies, and your own governance still has to explain how the tool is controlled.

What changed in April 2026

On April 17, 2026 the three federal banking agencies issued revised interagency guidance on model risk management:

  • Federal Reserve SR 26-2, "Revised Guidance on Model Risk Management." It supersedes SR 11-7, "Guidance on Model Risk Management" (April 4, 2011), and SR 21-8.
  • OCC Bulletin 2026-13, "Model Risk Management: Revised Guidance." It rescinds OCC Bulletin 2011-12, "Sound Practices for Model Risk Management: Supervisory Guidance on Model Risk Management," the "Model Risk Management" booklet of the Comptroller's Handbook, OCC Bulletin 1997-24 on credit scoring models, and OCC Bulletin 2021-19.
  • FDIC FIL-15-2026, "Agencies Revise the Interagency Model Risk Management Guidance." It rescinds FIL-22-2017 and FIL-27-2021.

Three points from the agencies' own pages shape any evaluation of AI lending software.

Size. The guidance is "expected to be most relevant to banking organizations with over $30 billion in total assets." The FDIC says it generally does not apply to models used by banking organizations of $30 billion or less, to the extent they lack significant model risk exposure. The OCC says the guidance is applicable to all community banks, subject to the limitations discussed in it. A community bank that buys AI underwriting software still has to explain to its examiner how it understands and controls the tool, even where the guidance is not the yardstick.

Generative and agentic AI. The OCC bulletin states: "Generative AI and agentic AI models are novel and rapidly evolving. As such, they are not within the scope of this guidance." The agencies said they plan a request for information on banks' use of AI. That leaves a gap for many AI lending tools, not a pass. Third-party risk management, your credit policy and your board's AI governance still apply to whatever the tool does.

Not enforceable on its own. The OCC and FDIC both say the guidance "does not set forth enforceable standards or prescriptive requirements." It describes sound practice, and your examiner will still ask how you manage the risk.

Is the software a model?

The revised guidance defines a model as "a complex quantitative method, system, or approach that applies statistical, economic, or financial theories." It excludes simple arithmetic calculations, such as those in spreadsheets, and deterministic rule-based processes where no such theory is applied.

AI lending software usually does several jobs, and they can fall on different sides of that line. Break the tool into its parts before you classify it:

  • Reading documents. Pulling figures out of rent rolls, operating statements and tax returns. Ask how the vendor's AI does it, and whether any of it is generative AI.
  • Calculating ratios. Coverage, loan to value and debt yield are arithmetic once the inputs are set. Ask whether the formula is shown and can be rerun by hand.
  • Checking policy. Testing a deal against your written credit policy may be rule-based. Ask how a policy is turned into tests and who signs off on them.
  • Scoring, forecasting or recommending. Any part that predicts performance, scores risk or drafts a recommendation deserves the closest look.

Record your classification and the reasons for it. Your model inventory, if you keep one, should say which parts are in scope and why.

What the guidance expects of vendor products

The revised guidance has a section on vendor and other third-party products. It recognizes that a bank "may not receive from the vendor the underlying code, data, or methodology," and still says "the principles of model risk management remain applicable." It points to validation of vendor products by internal or outside parties, to understanding the vendor model "including its conceptual soundness," and to assessing whether vendor models "are accurate, remain fit for purpose, and continue to be reliable."

Turn that into requests a vendor can answer in writing:

  1. A plain description. What the software does, what it does not do, and which outputs reach a credit decision.
  2. Decision rights. Written confirmation that your people approve every credit decision, and where in the workflow that happens.
  3. Model documentation. What your model risk reviewer can read about how outputs are produced and checked.
  4. Traceability. Whether every figure can be traced to the document and page it came from, so a reviewer can check an output against its source.
  5. Handling uncertainty. What the software does when it is unsure of a value: guess, flag, or hold it for a person.
  6. Outcomes analysis. The guidance describes comparing "model outputs to corresponding real-world outcomes." Ask to run the tool on closed files your team already decided and compare.
  7. Ongoing monitoring. How you will know the tool still performs as expected when your products, borrowers or documents change.
  8. Change control. How you are told when models or logic change, and how a change is tested before it reaches your deals.
  9. Accuracy claims. If a vendor quotes an accuracy figure, ask for the sample, the method and who measured it.

Third-party risk still governs the relationship

Whatever you conclude about model scope, the vendor is a third party. The interagency guidance on third-party relationships was issued in June 2023 as OCC Bulletin 2023-17, "Third-Party Relationships: Interagency Guidance on Risk Management" (June 6, 2023), and Federal Reserve SR 23-4, "Interagency Guidance on Third-Party Relationships: Risk Management" (June 7, 2023). The OCC bulletin states that the guidance applies to all banks with third-party relationships, and it describes a risk management life cycle for those relationships.

That guidance is under review. On September 11, 2026 the NCUA, FDIC, Federal Reserve Board and OCC requested comment on proposed third-party risk management guidance and said they plan to rescind their existing guidance once it is final. On November 13, 2026 they extended the comment period to December 16. Until a final version is issued, OCC Bulletin 2023-17 and SR 23-4 remain in place on the agencies' sites.

Credit unions

The interagency model risk guidance comes from the three bank agencies. For credit unions, NCUA's Letter to Credit Unions 07-CU-13, "Evaluating Third Party Relationships" (December 2007), sets out due diligence for third-party relationships and says the review should take into account the critical nature of the service. NCUA's 2026 supervisory priorities letter (26-CU-01, January 2026) says examiners will assess third-party risk-management practices when lending, servicing or collection functions are outsourced. The questions in this guide apply to a credit union's review of AI lending software in the same way.

How LenderBox answers these questions

We are a vendor, so here are our answers in the same order. LenderBox does not approve loans: it drafts the credit memo and the recommendation, and your people decide, with a human in the loop at every stage. Model documentation is available for your model risk team, and we walk the team through how LenderBox works and answer its questions. Every figure from the file is cited to its document and page, ratios are calculated by formula with the formula shown, and market figures are marked as market data from third-party data providers. A figure LenderBox is unsure of waits for your analyst, and the review cannot close until every one is confirmed. The demo can run on your own deal documents, so your team can compare the output with files it already decided. We make no accuracy guarantee and publish no accuracy percentage: we design for near-perfect accuracy, and a machine can still make mistakes, so a person stays in the loop for every judgment call. On the third-party side, LenderBox holds a SOC 2 Type II report from an independent auditor, released under NDA, and data is hosted on AWS in United States data centers, encrypted, isolated per institution and never used to train or improve any model. Our vendor due diligence page lists each item with the document behind it.

Frequently asked questions

Has SR 11-7 been replaced?

Yes. Federal Reserve SR 26-2, issued April 17, 2026, supersedes SR 11-7. The OCC rescinded OCC Bulletin 2011-12 the same day through OCC Bulletin 2026-13, and the FDIC issued FIL-15-2026.

Does the revised model risk guidance apply to a community bank?

The agencies say it is expected to be most relevant to banking organizations with over $30 billion in total assets. The FDIC says it generally does not apply to smaller institutions without significant model risk exposure; the OCC says it applies to community banks subject to its limitations. Your compliance team decides how it applies to you.

Does model risk guidance cover generative AI?

No. The OCC bulletin states that generative AI and agentic AI models are not within the scope of the revised guidance. Third-party risk management and your own AI governance still apply.

Is AI underwriting software a model?

Parts of it may be. The guidance defines a model as a complex quantitative method that applies statistical, economic or financial theories, and excludes simple arithmetic and deterministic rule-based processes. Classify each part of the tool separately.

What should we ask an AI lending vendor for?

A plain description of what the tool does, written decision rights, model documentation, traceability of every figure, how uncertain values are handled, results on your own closed files, monitoring and change control, and the basis of any accuracy claim.

Which guidance covers the vendor relationship itself?

The June 2023 interagency guidance on third-party relationships (OCC Bulletin 2023-17 and Federal Reserve SR 23-4). The agencies and NCUA proposed replacement guidance in September 2026, with comments due December 16. For credit unions, NCUA Letter 07-CU-13 covers third-party relationships.

Sources

  • Federal Reserve SR 26-2, Revised Guidance on Model Risk Management, April 17, 2026: federalreserve.gov, and the attached guidance, SR2602a1.pdf. Read November 17, 2026.
  • Federal Reserve SR 11-7, Guidance on Model Risk Management, April 4, 2011: federalreserve.gov. Read November 17, 2026.
  • OCC Bulletin 2026-13, Model Risk Management: Revised Guidance, April 17, 2026: occ.gov. Read November 17, 2026.
  • FDIC FIL-15-2026, Agencies Revise the Interagency Model Risk Management Guidance, April 17, 2026: fdic.gov. Read November 17, 2026.
  • OCC Bulletin 2023-17, Third-Party Relationships: Interagency Guidance on Risk Management, June 6, 2023: occ.gov. Read November 17, 2026.
  • Federal Reserve SR 23-4, Interagency Guidance on Third-Party Relationships: Risk Management, June 7, 2023: federalreserve.gov. Read November 17, 2026.
  • NCUA, Agencies Seek Comment on Proposed Third-Party Risk Management Guidance, September 11, 2026: ncua.gov; comment period extended November 13, 2026: ncua.gov. Read November 17, 2026.
  • NCUA Letter to Credit Unions 07-CU-13, Evaluating Third Party Relationships (December 2007), as named on ncua.gov; letter text: ncua.gov. Read November 17, 2026.
  • NCUA Letter to Credit Unions 26-CU-01, NCUA's 2026 Supervisory Priorities, January 2026: ncua.gov. Read November 17, 2026.

Want our answers in writing? Request the SOC 2 report through the trust center, or see it on your deal.

Related reading

See it work on one of your own deals.

A 30-minute demo can run on documents you bring. Every figure comes back tied to its source page, and your team still makes every credit decision.

See it in action
30 MINUTES. NO COMMITMENT. 90-DAY MONEY-BACK GUARANTEE.