AI credit scoring, underwriting, and adverse-action compliance — independently audited
Explanation rights, fair-lending obligations, and CFPB adverse-action requirements apply regardless of model complexity. "The algorithm is too complex to explain" is not a compliance defence.
AI systems in scope for Fintech
Every system below is covered in a standard iDharma engagement. Complex or multi-system deployments are scoped on request.
End-to-end audit of feature selection, proxy variable risk, explanation generation, and adverse-action notice compliance.
Automated or AI-assisted underwriting decisions — disparate impact testing, documentation, and MRM alignment.
Models using rent, utility, cash-flow, or behavioural data — proxy discrimination risk and FCRA/ECOA compliance.
Real-time fraud and anti-money-laundering models — accuracy benchmarking, false-positive rate by demographic, and explainability.
What we audit against
Every iDharma Fintech engagement maps simultaneously against the frameworks below — producing one gap register, not three separate reports.
Grants individuals the right to an explanation for AI-based credit decisions. Applies to any EU-market deployment regardless of where the lender is headquartered.
Require specific, accurate reasons for credit denial. AI-generated reason codes that do not correspond to actual model factors create regulatory and litigation exposure.
Prohibit discriminatory lending decisions — including indirect discrimination via proxy variables in AI models.
How an iDharma audit works in Fintech
We do not accept vendor documentation as evidence. We do not produce checkbox compliance reports. Every audit produces a named auditor, a cited methodology, and a straight answer on exactly where your AI stands — and what to fix first.
See how we work →We test every model feature for proxy discrimination against protected classes — not just the obvious proxies.
We evaluate adverse-action reason codes against the model's actual decision logic to verify they are accurate and legally compliant.
We independently reconstruct the model's decision boundary on a sample of your production data — vendor accuracy claims are not taken at face value.
We produce documentation structured to satisfy both EU AI Act technical file requirements and US model risk management (SR 11-7) expectations.
Explore other domains
Deploying AI in lending or credit?
Start with the free Risk Snapshot to understand your adverse-action and fair-lending exposure before a regulator does.