Industries

When a model denies a loan, someone has to explain why. AVAAS certifies that it can.

Credit is decided by models. Underwriting, approval, pricing, and limit decisions increasingly run through machine-learning systems that even their owners struggle to explain. Federal law does not accept the algorithm decided as an answer. AVAAS certifies the model at the point it acts on an applicant.

ECOA / Reg BCFPB Circular 2022-03CFPB Circular 2023-03Fair lendingSR 11-7California ADMT
Assess your credit models →
Where AI acts on a person

The decision point in lending

In lending, the AI acts when it approves or denies an application, sets a rate, assigns a limit, or flags an account. Each of those is an adverse action the moment it goes against the consumer.

Every adverse action carries a legal duty to give specific, accurate reasons, and an opaque model that cannot produce them is a compliance failure waiting to be found.

What keeps you exposed

What keeps lenders exposed

The black box

Your model denies a loan and cannot say why

ECOA requires the principal reasons for any adverse action. A complex model that cannot produce them does not meet the standard.

Fair lending

Disparate impact you never intended

A model can price or deny along protected lines through proxies. If you are not testing for it, an examiner will.

Model risk

Supervisors expect independent validation

SR 11-7 sets an expectation for model-risk management. Self-validation rarely satisfies a skeptical examiner.

This is already happening
The CFPB has stated, in Circular 2022-03 and again in 2023-03, that a complex or black-box model is no excuse for failing to give a borrower specific, accurate reasons for a denial.
CFPB Circular 2022-03 · CFPB Circular 2023-03
How AVAAS adds value

Evidence the model does what ECOA requires

Can the model give specific, accurate reasons for every adverse action?

AVAAS evaluates the system with causal attribution, so a denial traces to the features that drove it rather than a post-hoc approximation.

Does it produce disparate impact across protected classes?

Five structurally independent validators test for demographic disparity in approvals and pricing.

Does independent validation exceed the SR 11-7 bar?

AVAAS is operated separately from the standard it certifies, so the validation is genuinely independent of the model owner.

You get documented, third-party evidence that the credit model does what the law requires, with sealed deployment verification confirming the certified model is the one in production.

Related AVAAS coverage: SR 11-7 · California ADMT · GDPR Article 22. Or run the free Regulation Checker to see what applies to you.

See where your lending AI creates liability.

Tell us where AI touches your underwriting, pricing, or account decisions, and we will scope an AVAAS certification to the exposure.

Ready to start now? Certify Your AI →  or  email [email protected]