Detect Bias and Discrimination in AI Approval Decisions
Compliance teams testing AI approval systems for discriminatory patterns before regulators find them.
El problema
Corules does not produce biased decisions because it evaluates against explicit, auditable policy criteria rather than learned model weights. However, if the policy itself encodes bias (e.g., a parameter that inadvertently correlates with protected class), Corules's audit log enables disparate impact analysis. Every decision is logged with the specific factors evaluated. Segmenting outcomes by protected class attributes reveals whether the policy produces disparate outcomes — and which specific rule is responsible.
Cómo lo resuelve Corules
Corules's policy runtime evaluates structured context against compiled CEL expressions — returning ALLOW, BLOCK, or ESCALATE with a reason and audit ID.
Ejemplo de política
// Audit log enables disparate impact analysis:
// Query: outcomes by applicant_zip_code for credit decisions
// If approval rates differ significantly by zip → investigate
// Corules shows WHICH factor caused each denial:
// "violation": "dti_ratio > params.max_dti"
// → Is max_dti set differently by geography? That would be the issue.Frequently Asked Questions
Does Corules prevent discriminatory outcomes?
Corules enforces policy deterministically. It does not evaluate protected class attributes unless the policy explicitly includes them. The audit log enables you to detect if policy-compliant decisions produce disparate outcomes.
Is this ECOA/FCRA compliant?
Corules produces specific adverse action reasons for every denial — a requirement under ECOA and FCRA. The audit trail is suitable for regulatory examination.