The Reserve Bank of India has introduced a new Model Risk Management framework requiring human supervision for all AI-driven banking decisions. Banks must now implement 'kill switches' and ensure human intervention is possible for automated processes. This move aims to prevent systemic risks as lenders increasingly adopt artificial intelligence in areas like lending and fraud detection.
The Reserve Bank of India (RBI) has released a draft Model Risk Management framework that places human control at the center of banking operations using artificial intelligence (AI) and machine learning (ML). As financial institutions in India move toward faster digital processes, the regulator is establishing clear rules to ensure banks remain accountable for automated decisions that impact customers and systemic stability.
Accountability in Automated Banking
Under the proposed guidelines, banks are required to maintain full responsibility for the performance of their analytical models, whether those systems are built by in-house teams or purchased from third-party technology providers. The framework mandates that banks create a board-approved policy covering the entire lifecycle of a model, from development to retirement. Every model must undergo independent validation before it goes live, even if the technology supplier has already certified it.
Operational Safeguards and Human Control
To address the risk of over-reliance on technology, the RBI has proposed the 'human-in-the-loop' principle. This ensures that bank employees retain the power to question, override, or stop automated decisions. In scenarios where a system produces unexpected results, banks will be required to activate emergency protocols, often called 'kill switches,' to immediately suspend the AI model.
The regulator has specifically highlighted concerns regarding automation bias, where staff might blindly trust computer outputs, and decision fatigue, which can lead to errors in supervision. For systems that interact directly with customers, the draft mandates that banks clearly disclose the use of AI and provide an option for customers to reach a human representative.
Governance and Risk Classification
Banks will be required to categorize their models based on complexity and risk level. Models identified as high-risk will need approval from a bank's risk management committee before they can be deployed. Additionally, institutions must maintain a detailed inventory of all models. Records for retired or decommissioned models must be kept for at least 10 years to allow for historical auditing and performance reviews.
Impact on Digital Banking Strategy
This framework marks a transition for the Indian banking sector, moving away from simple compliance toward building operational resilience. As banks integrate AI into sensitive functions such as credit underwriting, treasury management, and fraud detection, the cost of technology implementation may increase due to the requirement for specialized technical staff and independent validation processes. Investors may monitor how large private and public sector banks adjust their digital budgets and vendor partnerships to align with these stringent regulatory requirements in the coming quarters.