What does RBI draft circular on Regulatory Principles for Management of Model Risks in Credit say about AI in banking?

What are the key points of RBI draft circular on Regulatory Principles for Management of Model Risks in Credit?

• Definition (paragraph 1): a credit risk model is any quantitative method that applies statistical, economic, financial or mathematical principles and assumptions to process data into an output used for credit decisions, including credit scoring, borrower selection, loan pricing, risk analysis and loan loss provisions.
• Governance (paragraphs 2 and 3): a detailed Board-approved policy covering the entire model lifecycle, a model inventory of approved models whether in-sourced or outsourced, and RMCB (or designated Board sub-committee) approval of each credit model and of subsequent changes in inputs or assumptions.
• Development and deployment (paragraph 4): defined objectives, robust inputs and assumptions, detailed documentation of output sensitivity, and outcomes that are consistent, unbiased, explainable and verifiable; overrides based on subjective factors must follow the policy and be documented in an auditable format.
• Third-party models (paragraph 5): the same principles apply; contracts must give access to minimum technical documentation, and REs remain ultimately responsible and accountable for the integrity and outcomes of outsourced models.
• Validation (paragraphs 6 to 8): independent validation before deployment, after amendments owing to material events and at least yearly, covering assumptions, data accuracy, regulatory compliance, documentation, back-testing and instances of bias or discrimination; results go to the RMCB.
• Supervisory review (paragraph 9): deployed models are subject to supervisory review, and the RBI may engage external experts to validate models, including external models, with contractual provisions enabling supervisory evaluation.
• Timeline in the draft (paragraph 3 of the covering circular): in force within three months of issuance; new models follow it from then, existing models to be validated within six months.
• Repeal clause: the draft would repeal Chapter 3 (Credit Risk Models) alone of the Guidance Note on Credit Risk Management dated 12 October 2002.

What did RBI draft circular on Regulatory Principles for Management of Model Risks in Credit change for banks?

It proposed binding model risk principles for credit models, aimed at borrower selection, scoring and pricing, and it already required explainable and unbiased outcomes without naming AI. It was overtaken in scope by the June 2026 draft Guidance on Regulatory Principles for Model Risk Management, which covers all models and adds AI/ML principles.

What did the RBI's draft Regulatory Principles for Management of Model Risks in Credit propose for banks?

The RBI's draft circular of 5 August 2024 proposed that regulated entities use a Board-approved policy for the full lifecycle of credit models, keep an inventory of all approved models, whether in-sourced or outsourced, and obtain approval from the Risk Management Committee of the Board for each credit model and any change in its inputs or assumptions. Each model would be independently validated before deployment, after material amendments and at least yearly, with back-testing and review for bias or discrimination, and third-party models would be subject to the same principles with contractual access to documentation and supervisory evaluation. It was published for comment until 4 September 2024, was drafted to apply three months after issuance with existing models validated within six months, and has not been issued in final form; the June 2026 draft Guidance on Regulatory Principles for Model Risk Management is the RBI's later, broader proposal.

The draft's covering circular would have brought the principles into force within three months of issuance, with existing models validated within six months, and would have repealed Chapter 3 on Credit Risk Models of the 2002 Guidance Note on Credit Risk Management. Neither has happened because the circular was not finalised.

The RBI's press release of 24 June 2026 describes the new draft Guidance as following the 2024 draft and the FREE-AI report, and the new draft's paragraph 64 provides for it to supersede the same Chapter 3 on finalisation. Banks should therefore treat the 2024 text as background and plan against the 2026 draft.

WHAT THIS MEANS IN PRACTICE

• Use the 2024 draft's credit-model checklist (inventory, RMCB approval, annual validation, bias review) as a baseline, since the 2026 draft retains the same architecture for all models.
• Check that credit models sourced from fintech or collaborative lending partners have contractual documentation access and supervisory evaluation rights.
• Document any manual overrides of credit model outcomes in an auditable format.
• Do not treat the 2024 text as binding; follow the final guidance when the RBI issues it.