The Reserve Bank of India has circulated draft guidelines that would force banks to strengthen governance and controls around artificial intelligence (AI) and machine-learning models.
Under the proposal, every regulated entity would need a risk management framework approved by its board that applies to the full set of models in use - explicitly including AI and machine-learning applications. That framework must encompass both individual model risks and risks seen across the institution as a whole.
Ongoing assessments are central to the draft. Banks would be required to run continuous risk evaluations at the level of each model and to aggregate those findings to understand enterprise-wide exposure. Where assessments reveal excessive risk, the guidelines specify that banks must take corrective steps. Acceptable responses listed in the draft include the introduction of enhanced controls, placing restrictions on model usage, remediating the model or decommissioning it entirely. Following any significant corrective action, a report is to be furnished to the board's risk management committee.
The draft also mandates independent validation for all models, covering systems developed in-house as well as those supplied by third parties. In addition, institutions must maintain inventories cataloguing the models they use.
Human oversight is required for AI systems deployed in automated decision-making roles, according to the document. For generative AI models that engage with customers or external users, the Reserve Bank says banks must implement additional cybersecurity controls to guard against risks introduced by those interactions.
The central bank has opened the draft guidelines for consultation and has invited feedback by July 24.
What this means in practice
The draft sets out clear expectations for board-level governance, continuous monitoring, independent validation and specific operational safeguards for AI-driven systems. It also highlights an explicit focus on third-party models and on human oversight where automated decisions affect customers.