AI Regulation Compliance

Explanation

AI Regulation Compliance checks whether an AI system used in banking actually satisfies the laws and supervisory rules that apply to it — for example RBI accountability expectations and fair-lending obligations on a credit-decision model — so that the bank does not deploy automation that is unlawful or unauthorised. It matters because a single non-compliant high-risk model (say a loan-approval engine that ignores fair-lending requirements) can trigger penalties, customer harm and reputational damage. It is measured by the Regulatory Alignment Score, computed as Alignment Score = (Σ(Requirement_Met_i × Requirement_Weight_i)) / (Σ(Requirement_Weight_i)); for lending decisions this is alignment with fair lending laws, so each applicable requirement is weighted by importance and scored as met or not. To implement it, maintain a structured register of legal/regulatory requirements (using document-extraction and RAG over the rulebooks) mapped to specific system controls, assign weights and ownership per NIST GOVERN 1.1/2.1 and ISO Clause 5.2/5.3, evaluate each requirement at design and before release, and log the evidence (who assessed, against which clause, with what result). The threshold is 100% compliance for high-risk applications; any shortfall there is an immediate-priority breach that must block deployment until every weighted requirement is met.