Industries · Banking & Lending

Prove every lending decision was fair — before an examiner asks.

EVE CoreGuard enforces your fair-lending policy — ECOA, Reg B, FCRA, SR 11-7 — on each AI-assisted lending decision before the model’s output reaches an applicant, and signs a replayable evidence record that a regulator can verify offline. Block the decision you can’t defend.

Policy decision in <1ms  ·  Signed, hash-chained record  ·  Offline-replayable for examiners
Enforces against ECOA / Reg B FCRA SR 11-7 Fair Housing Act UDAAP
The exam question

A model card won’t answer “prove this AI decision was compliant.”

AI is moving into credit underwriting, pricing, fraud, and collections. The problem isn’t whether the model is accurate — it’s whether you can prove, decision by decision, that it stayed inside the law. A probabilistic model plus application logs cannot show an examiner why a specific denial was issued, that the adverse-action reasons were present, or that a model update didn’t introduce disparate impact.

ECOA · Reg B · FCRA

Adverse-action exposure

A denial that lacks the specific principal reasons required by Reg B (12 CFR 1002.9) and FCRA is a violation per notice — and these scale fast across a portfolio.

ECOA · Fair Housing Act

Disparate impact on model updates

A retrained or re-tuned model can quietly shift outcomes for a protected class. Without pre-promotion testing, the first time you learn is during an exam or an investigation.

SR 11-7

Model risk documentation

Examiners expect controls that are consistent, explainable, and auditable. Probabilistic outputs that drift with model version are none of those.

UDAAP

Inconsistent decisions

If the same applicant profile can get different outcomes across model versions, you can’t explain — or defend — why a given decision happened.

What EVE CoreGuard does

Deterministic enforcement, then signed evidence — on every decision.

CoreGuard sits in front of your lending model as a governance layer. It evaluates each proposed decision against your lending_v1 policy pack and returns ALLOW, BLOCK, or MODIFY before the output is finalized — then writes a cryptographically signed record of exactly which rule fired and why.

1

Enforce policy before the decision is final

The lending_v1 pack encodes ECOA / Reg B / FCRA / SR 11-7 rules. The same input always produces the same governance decision — deterministic, not probabilistic.

2

Block adverse actions that lack required reasons

If a denial is missing the principal reason codes Reg B and FCRA require, CoreGuard blocks it and records the gap — so the notice you send matches the record you keep.

3

Gate every model update through the Model Update Firewall

Each model change is simulated against your fair-lending rules and blocked before promotion if it introduces disparate impact. See the EVE Model Update Firewall →

4

Hand the examiner a record they can verify themselves

Every decision becomes a signed, hash-chained evidence record (Ed25519 in production). Re-hash and re-verify it offline with the public key — no EVE service required. Verify a record →

governed decision · signed evidence record ✓ VERIFIED
decision_idDEC-00042
policylending_v1 · ECOA / Reg B
verdictBLOCK — adverse-action reasons required
content_hashsha256:3204f3d6…1ef0f3130
signatureed25519:4e542efc…a10250b02
Sample record · re-hash + Ed25519 re-verify, no EVE service required Verify a record offline →
The economics

One prevented fair-lending event pays for years of governance.

The price tag on a single lending-governance failure dwarfs the cost of the control that prevents it.

Governance failureIllustrative costWhat drives the number
Fair-lending investigation or settlementECOA / FHA · redlining or disparate impact
$2M–$30M+
Public DOJ / CFPB consent orders pair a civil penalty with a consumer remediation fund.
Adverse-action notice failuresECOA / Reg B · FCRA
$100K–$1M+
FCRA statutory damages run $100–$1,000 per affected consumer (15 U.S.C. §1681n) and scale with class size.
Failed exam → MRA remediationCompliance / safety-and-soundness finding
$500K–$2M+
Lookback review, outside consultants, and added control staffing to clear a Matter Requiring Attention.

Illustrative ranges drawn from public regulatory penalty caps, published enforcement actions, and statutory damages — not EVE customer results. Model your own exposure with the ROI calculator. EVE CoreGuard’s Enforcement license is $150,000/year.

Deployment

Your data never leaves your tenant.

CoreGuard runs as SaaS, in your VPC, or fully on-prem. The governance decision and the signed record are produced inside your boundary — nothing about an applicant is sent to EVE to make a decision. See deployment models →

SaaS

Fastest start. Decisions and signed records produced in an isolated tenant.

VPC / Private

Runs inside your cloud account, under your network and key controls.

On-prem

Air-gap-friendly for institutions that keep model decisioning fully in-house.

Examiner access

Issue scoped, time-boxed evidence links so an examiner can verify records directly.

Questions buyers ask

Fair-lending governance, answered plainly.

No. Your model and your underwriters make the decision. CoreGuard governs it: it evaluates the proposed action against your lending policy pack before the output is finalized, returns ALLOW / BLOCK / MODIFY, and produces a signed evidence record. It is a governance and evidence layer, not a credit model.
When an AI-assisted decision results in a denial, Reg B (12 CFR 1002.9) and FCRA require specific principal reasons. CoreGuard can block a denial that lacks the required reason codes and record the reasons that were present — so the notice you send and the record you keep are consistent and traceable to the rule that fired.
SR 11-7 expects controls that are consistent, explainable, and auditable. CoreGuard applies deterministic, versioned policy — the same input always produces the same governance decision — and writes a signed, hash-chained audit trail. The Model Update Firewall additionally simulates every model update against your fair-lending rules before it is promoted.
Yes. Each record carries a content hash and a cryptographic signature (Ed25519 in production). An examiner can re-hash the record and re-verify the signature offline with the public key, no access to EVE’s service required. Verification proves the record was not altered after the decision was made.
Engagements start with a $37,500 design-partner pilot, scoped to a single lending workflow, with the pilot fee credited toward an annual license. The Enforcement license is $150,000/year. See the pricing page for current tiers.

Bring one lending workflow under deterministic governance.

A 60-day design-partner pilot puts CoreGuard in front of a single decision flow, produces signed evidence on real decisions, and credits the pilot fee toward your license.

Built by a lending-technology founder. 90 U.S. patent applications filed.