AI Regulation Tracker / Draft guidance

India's Central Bank Wants a Kill Switch on Every Bank AI Model

Regulatory summary: The Reserve Bank of India released its Draft Guidance on Regulatory Principles for Model Risk Management, 2026 on June 24, 2026, and opened it for public consultation until July 24, 2026. It is the RBI's first institution-wide framework covering all analytical models, including AI and machine learning, used by regulated.

The RBI's draft Model Risk Management guidance would force every bank, NBFC, and financial institution to govern all analytical models, AI included, under one board-owned framework, with a comment window closing July 24, 2026.

Primary source

By Anthony Guerriero, Founder, The Leveraged Years · Reviewed by The Leveraged Years Editorial Desk · Published July 9, 2026 · Last updated July 9, 2026

India's Central Bank Wants a Kill Switch on Every Bank AI Model regulation briefing

The Leveraged Years AI Regulation Tracker

Key takeaways

For the first time, the RBI proposes a single, institution-wide model-risk regime rather than model rules scattered across credit, market, and capital circulars. AI and ML models are pulled inside that regime and given extra obligations: kill-switch arrangements, human oversight, anti-automation-bias controls, and customer-facing AI disclosure.
Model-risk and validation teams, chief risk officers, boards and risk management committees, compliance officers, fintech and AI vendors selling into Indian financial institutions, and financial-services counsel advising on AI deployment.
Status: Draft.
Inventory every analytical model now, including shadow and vendor models, and map each against the draft's controls to size the gap. Firms with a view should file comments before July 24, 2026.

Regulatory Briefing

Bindingness: Draft supervisory guidance (public consultation). Not binding in its current form.
Scope: RBI-regulated entities, including commercial banks, small finance banks, payments banks, local area banks, regional rural banks, urban and rural cooperative banks, NBFCs across all layers (which includes housing finance companies as a category of NBFC), all-India financial institutions (NABARD, SIDBI, NHB, NaBFID, EXIM Bank), asset reconstruction companies, and credit information companies. The framework governs all analytical models, with AI and machine learning systems singled out for additional controls.
What Changed: For the first time, the RBI proposes a single, institution-wide model-risk regime rather than model rules scattered across credit, market, and capital circulars. AI and ML models are pulled inside that regime and given extra obligations: kill-switch arrangements, human oversight, anti-automation-bias controls, and customer-facing AI disclosure.
Who This Affects: Model-risk and validation teams, chief risk officers, boards and risk management committees, compliance officers, fintech and AI vendors selling into Indian financial institutions, and financial-services counsel advising on AI deployment.
Current Status: Draft. Comment period open to July 24, 2026.
Effective Date: Not yet in force. This is a draft released for consultation; no implementation date is set.
Primary Source: https://www.rbi.org.in/Scripts/bs_viewcontent.aspx?Id=5089
Secondary Analysis: Coverage from Business Standard, MediaNama, and industry commentary frames the draft as bringing AI model governance to parity with traditional model risk and flags the kill-switch mandate as its most striking feature.
Workflow Impact: Institutions would need to stand up or upgrade a board-approved framework, build a defensible model inventory covering active and retired models, define a tiering methodology, and retrofit human-control and disclosure features into deployed AI. Independent RE validation of vendor models becomes non-negotiable regardless of vendor certification.
Operational Next Step: Inventory every analytical model now, including shadow and vendor models, and map each against the draft's controls to size the gap. Firms with a view should file comments before July 24, 2026.
Last Verified: July 9, 2026

Date	Jurisdiction	Rule	Affected professionals	Status or effective date
2026-07-09	India	For the first time, the RBI proposes a single, institution-wide model-risk regime rather than model rules scattered across credit, market, and capital circulars. AI and ML models are pulled inside that regime and given extra obligations: kill-switch arrangements, human oversight, anti-automation-bias controls, and customer-facing AI disclosure.	Model-risk and validation teams, chief risk officers, boards and risk management committees, compliance officers, fintech and AI vendors selling into Indian financial institutions, and financial-services counsel advising on AI deployment.	Draft. Comment period open to July 24, 2026.

Frequently Asked Questions

Is the RBI's Model Risk Management guidance in force?

No. It is a draft released on June 24, 2026 for public consultation until July 24, 2026. It carries no binding obligation in its current form, and no implementation date has been set.

Does the kill-switch requirement apply to all models or only AI?

The kill-switch, human-in-the-loop, human-override, and anti-automation-bias controls are directed at AI and ML systems specifically. The broader framework, including tiering, inventory, and validation, applies to all analytical models.

Does a vendor's certification satisfy the validation requirement?

No. The draft requires the regulated entity to independently validate all models, including third-party and vendor models, regardless of any certification the vendor provides.

How long must retired models be kept in the inventory?

At least ten years from the date of decommissioning, or from the date the model ceases to serve as a backup or benchmark reference, whichever is later.

Who is in scope?

Commercial banks, small finance banks, payments banks, local area banks, regional rural banks, cooperative banks, NBFCs across all layers, all-India financial institutions, asset reconstruction companies, and credit information companies.