RBI Proposes AI Model-Risk Rules With a Mandatory Kill Switch

The Reserve Bank of India has released draft Guidance on Regulatory Principles for Model Risk Management, 2026, moving to place the growing use of artificial intelligence and automated models under formal governance. As banks, non-bank lenders, cooperative banks, asset reconstruction companies and credit information companies increasingly deploy models in areas such as credit scoring, fraud detection, collections, customer service and risk pricing, the draft sets out expectations designed to ensure these tools are used safely and fairly. Key requirements include board-approved model risk frameworks, meaningful human oversight, independent validation of models, controls over third-party and vendor models, disclosure to customers where AI materially influences decisions, and a 'kill switch' that allows institutions to quickly deactivate models that pose a risk. Public comments are open until July 24, 2026. Regulators note that a flawed model can unfairly deny credit, misclassify borrowers, introduce bias or trigger wrong recovery actions, and can expose institutions to cyber and reputational risk. The framework aims to protect fairness and accountability, though smaller entities may need support and skilled teams to implement it without turning compliance into a box-ticking exercise.