RBI Mandates AI Kill Switch for Banks, Releases Draft Model Risk Management Framework

Key Takeaways

• On June 24, 2026, the RBI released its draft “Guidance on Regulatory Principles for Model Risk Management, 2026,” mandating banks build kill-switch arrangements into all AI systems.
   
• This builds on RBI's earlier draft from August 2024 on credit model risks, and the FREE-AI Committee report published in August 2025, now extended to cover all models.

Why Has the RBI Mandated a Kill Switch for AI Models Used by Banks?

On June 24, 2026, the RBI released the draft via Press Release 2026-2027/528, signed by Brij Raj, Chief General Manager, Department of Regulation. The draft applies to 11 categories of regulated entities, including commercial banks, small finance banks, payments banks, NBFCs, and credit information companies. Public comments are invited until July 24, 2026.

The RBI stated, “An RE (regulated entity) is accountable for the outcomes of all models used by it, irrespective of whether the models are developed internally, sourced from third parties, or a combination thereof.” The guidance requires institutions to establish “override, suspension, or deactivation mechanisms, including kill-switch arrangements,” so AI systems can be halted immediately if they produce problematic outcomes. The RBI also flagged “automation bias,” the tendency of employees to accept AI outputs without independent judgment.

How Will This RBI Framework Affect Everyday Bank Customers and Loan Applicants?

For customer-facing AI systems, banks must disclose when customers are interacting with an AI model and must let them switch to a human “at any point.” This directly affects how loan approvals, credit checks, and customer service chatbots will operate going forward. The framework requires banks to define “explainability and transparency thresholds” for every AI model, so loan rejections and approvals can be explained in plain terms.

AI-driven lending has already changed how fast loans get approved in India. LoansJagat documents a case where a borrower applied online at 10 AM, and an AI engine processed her 3 years of banking history and ₹50,000 in monthly transactions within

What Do Industry Experts Say About the RBI’s New AI Governance Rules?

The RBI introduced a three-lines-of-defence structure, with “model owners acting as the first line of defence, independent validation functions serving as the second, and internal audit providing the third layer of oversight.” The framework also flagged “supply chain risk,” the danger of banks over-relying on a small number of global AI providers.

Analysts at The420.in noted the framework specifically targets “model drift,” the gradual degradation of an algorithm's accuracy as market conditions diverge from its training data. The RBI’s solution requires “continuous validation loops and stress-testing protocols” to catch this drift early, alongside mandatory deactivation protocols that remain “completely accessible” to risk teams even if an AI model affects core banking systems.

Conclusion

The RBI's June 24, 2026 draft framework marks India's first comprehensive, board-level accountability structure for AI risk across the entire financial sector. With comments due by July 24, 2026, banks now have a clear runway to build kill switches and explainability standards into their fast-growing AI lending systems.

FAQs

What is the exact nature of the kill switch requirement from the RBI for AI in banks, and what is the effective date for the same?

The RBI issued guidelines on June 24, 2026, mandating all banks and regulated financial organisations to install a mandatory kill switch that will immediately turn off any AI program causing any negative effects. The period for receiving public comments ends on July 24, 2026, for all 11 types of regulated institutions.

What are the 3 biggest governance changes Indian banks must make once AI becomes core to their operations?

First, high-risk AI models will need explicit approval from the Board's Risk Management Committee. Second, banks remain fully responsible for the third-party AI models they deploy. Third, customers must be informed when AI influences a decision and given an option to speak to a human.