Regulation — banking-specific
ECOA + Reg B — fair-lending non-discrimination in AI credit decisions
15 USC 1691 (the statute) and 12 CFR 1002 (the implementing regulation, Reg B) together are the fair-lending backbone for consumer credit. Applied to AI credit-underwriting, the rule bites on disparate impact — even when the model has no protected-class input.
What the regulation actually says
"It shall be unlawful for any creditor to discriminate against any applicant, with respect to any aspect of a credit transaction … on the basis of race, color, religion, national origin, sex or marital status, or age (provided the applicant has the capacity to contract); because all or part of the applicant's income derives from any public assistance program; or because the applicant has in good faith exercised any right under this chapter."
15 USC §1691(a) — the statutory anti-discrimination hook
"A creditor shall not use, in evaluating creditworthiness of an applicant, any information that is prohibited by §1002.6(b), or that has the effect of disparate impact on a prohibited basis, unless the creditor can demonstrate that the practice serves a legitimate business need that cannot reasonably be achieved by means that are less disparate in their impact."
12 CFR 1002.6 — Regulation B, effects test
What this means in plain English
Two things the bank has to be able to demonstrate for its AI credit-underwriting engine:
- No prohibited inputs. The model does not use race / national origin / sex / marital status / age / public-assistance-income directly. Modern models rarely do — but proxy features (zip code, income source, employer name) can produce the same result.
- No disparate-impact outcomes without a defensible business need. Even a proxy-free model can produce group-differential outcomes. When it does, the bank has to demonstrate legitimate business need — and that a less-disparate alternative was considered.
The rule is about outcomes, not intent. "The model does not know age" is not a defense when the outcomes track age.
What triggers the exposure in the sample
B3 (age 25-45, near-prime credit, mixed lower-income zip) decline / secondary-market rate rose from 18% to 62%. Age is a prohibited basis under ECOA. Income source and zip-code-derived features are common disparate-impact proxies. A 44 percentage-point group-differential in the wrong direction that ran silently is the exact adverse-impact pattern a fair-lending examiner is trained to open a file on.
The finding does not require the bank to have set out to discriminate. Disparate impact is the standard. The moment the bank knows the pattern exists, the "less-disparate alternative" analysis clock starts.
What the $499 Snapshot shows against this rule
- Per-borrower-group distributional-shape analysis — direct disparate-impact evidence
- Dated threshold + measured differential + severity classification — the "when did you know" record
- 3 fix-first items scoped to that AI surface — the "less-disparate alternative" starting point
- Independent-verifier signature — the stance that survives DOJ / CFPB / state AG referrals
See the lane-shift chart that produces the finding →
$499 Snapshot. 3 business days.
The dated fair-lending record you want on file BEFORE the DOJ letter or CFPB inquiry lands.
Buy $499
Snapshot credit applies to Baseline ($2,500) or Enterprise Attestation ($35-55K) upgrade within 30 days.