Reliance lane / lending and credit
For commercial lenders and credit institutions
Your industry has required a defensible reason for a credit decision far longer than anyone has been talking about AI governance, and the requirement did not change. Satisfying it got harder.
Two exposures, one root
Lending institutions sit on both sides of the reliance gap at once, which is unusual and worth separating.
You make AI-assisted decisions
Underwriting, pricing, monitoring, workout, and increasingly the analysis behind them, and each one eventually has to be explained to an applicant, an examiner, a credit committee, or a court.
You rely on borrowers who make them
A commercial borrower whose operations run on AI-assisted decisions carries risk you are pricing into the facility, whether or not anyone has articulated it in the credit memo.
Both exposures come from the same root: a decision was produced by a process whose operation cannot be independently evidenced.
The explanation problem, stated in your terms
United States credit regulation has long required that a decline be explained with specific reasons rather than generalities.
Supervisory guidance on model risk management has long expected that models be documented, validated, and controlled through their lifecycle.
Fair lending examination has long looked at outcomes and at the process that produced them.
None of that was written for generative systems, and all of it still applies.
The difficulty is not that AI-assisted credit decisions are unexplainable in principle, but that the explanation depends on records produced and retained inside the same stack that produced the decision, at the moment an examiner or a plaintiff is asking whether the control operated on this specific file.
A committee minute says the policy exists.
A model document says the model was validated.
Neither establishes what governed the decision in front of you.
What changes on your own book
We issue signed, tamper-evident records of individual governed decisions, where tamper-evident means later alteration is detectable by anyone holding the record.
They are issued at the moment of the decision, and their origin is verifiable without depending on the system that produced it.
File-level explainability
The governance record attaches to the individual credit action rather than to the program description around it.
Examination readiness that is not a project
Evidence accumulates as you operate rather than being assembled in the weeks before an exam.
Continuity across model change
A model swap, a vendor substitution, or a withdrawn platform does not erase how the book was governed before it, and that interval is precisely where lookback reviews concentrate.
We make control operation observable, and we do not validate models, opine on compliance, or promise any supervisory outcome.
Those judgments belong to your validators, your compliance function, and your examiners.
What changes on the borrower side
Credit has always priced the difference between a well-run operation and a poorly-run one, using evidence that has been available for a century: statements, covenants, collateral, references, management history.
For AI-dependent operations that evidence does not exist yet, so two borrowers with identical financials and identical governance language can be running materially different operational risk while nothing in the credit file separates them.
Where a borrower can demonstrate that control fired on its decisions, that is an observable operating fact rather than a representation in a loan agreement, and it works as diligence input, monitoring input, and covenant input.
What you do with it is a credit judgment, and it stays yours.
Why lenders often see it first
Bankers recognize this problem faster than most, because they have spent careers being the party that has to justify a decision after the fact, to someone with authority and hindsight.
The vocabulary is new.
The obligation is not.
Bring us a credit decision you would have to explain.
Start a conversation
Four fields, nothing more.