As lenders bring more data and AI into credit decisions, borrowers need a way to challenge mistakes. A PYMNTS report says bank examiners are being told to assess AI according to its purpose and the authority it has, while lenders must be able to explain decisions that affect applicants.
Watch Desk analysis
What happened
The report describes Federal Reserve Governor Michael Barr’s call for accessible dispute procedures, ongoing fairness monitoring and detailed records of how AI models are validated. It says the Office of the Comptroller of the Currency’s 4 November framework tells examiners to tailor reviews to a system’s potential consequences and autonomy.
There is a practical distinction for borrowers. When a consumer reporting agency supplied information used in a credit denial, the lender must identify that company. If a landlord supplied rental information directly, the borrower generally has the right to request a description of it from the lender. The report says lenders must explain their decision within 60 days of a written request.
Why it matters
Using rental payments, utility bills or employment patterns could help lenders assess people with little conventional credit history. But broader data only helps if applicants can see what was used, correct errors and understand how those details shaped a decision. A model’s explanation is not much use if nobody can retrieve the data or model version behind the original call.
Our read
The useful test for AI in lending is not simply whether it can process more information. It is whether the bank can show what counted, fix what was wrong and explain the outcome to the person affected. “The model decided” is not an explanation; it is a very expensive shrug.
What to watch
- Whether banks keep records of the data and model version used for individual decisions.
- How examiners apply the OCC’s framework to systems with different levels of autonomy.
- Whether borrowers can challenge errors in non-traditional financial data in practice.
Discussion spark: Should lenders have to disclose the data and model logic behind every AI-assisted credit denial, or would that make fair lending harder to manage?
Sources and evidence
- Source update (18 November 2026, 17:52 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.