A human firmly in the loop: RBI proposes ‘kill switch’ for AI used by banks & NBFCs

The Reserve Bank of India (RBI) released draft "Guidance on Regulatory Principles for Model Risk Management" (PR 2026-2027/528) on June 24, 2026, introducing...

What Happened

The Reserve Bank of India (RBI) released draft "Guidance on Regulatory Principles for Model Risk Management" (PR 2026-2027/528) on June 24, 2026, introducing a comprehensive framework for the governance of AI and machine learning models used by regulated financial entities.
The draft framework mandates a "kill switch" — a mechanism to immediately override, suspend, or deactivate any AI/ML model if it produces harmful, biased, or erroneous outputs.
Responsibility for every AI-driven decision is placed squarely with the board of the regulated entity; financial institutions cannot transfer liability to third-party technology vendors.
The framework applies to commercial banks, small finance banks, payments banks, local area banks, co-operative banks, regional rural banks, NBFCs, all-India financial institutions (NABARD, NHB, EXIM, SIDBI), asset reconstruction companies, and credit information companies.
Public comments on the draft have been invited until July 24, 2026.

Static Topic Bridges

Model Risk Management Framework (MRMF) and AI Governance in Finance

A Model Risk Management Framework (MRMF) is a structured governance system that covers the entire lifecycle of quantitative and AI/ML models used in financial decision-making — from design and validation through deployment, monitoring, and eventual decommissioning. As financial institutions use AI for credit scoring, fraud detection, loan underwriting, and customer-facing chatbots, the risk that a flawed model causes large-scale harm has increased significantly.

The RBI's draft MRMF introduces "risk-based model tiering" — classifying models by the complexity of their decision impact, with higher-tier models (e.g., credit decisioning AI) subject to stricter validation and oversight.
Key requirements: board-approved MRMF policy; explainability thresholds (AI decisions must be interpretable); safeguards against model hallucinations; testing under adversarial and abnormal scenarios; bias and discrimination monitoring; and continuous model behaviour assessment.
No "black box" AI: institutions must be able to explain the basis of any AI-driven financial decision to customers and regulators.
Third-party accountability: banks cannot outsource accountability — an institution using a vendor-supplied AI model is fully responsible for its outcomes.

Connection to this news: The kill switch mandate directly implements the principle of human-in-the-loop governance: even highly automated AI systems must have a clearly defined path by which a human can halt operations, creating a minimum standard of accountability across all AI deployments in Indian finance.

RBI's Regulatory Architecture and Powers

The Reserve Bank of India was established under the Reserve Bank of India Act, 1934 and is the central bank and principal financial sector regulator. Its supervisory authority over banks derives from the Banking Regulation Act, 1949; its authority over NBFCs derives from Chapter III-B of the RBI Act, 1934.

RBI is a statutory regulator with powers to issue binding directions, impose penalties, cancel licences, and appoint administrators for defaulting entities.
Draft guidelines issued by RBI for public consultation are typically followed by final circulars/master directions that become legally binding on regulated entities.
RBI's Department of Regulation (DoR) oversees prudential norms; the Department of Supervision (DoS) oversees compliance and risk management.
RBI operates under the broader framework of the Financial Stability and Development Council (FSDC) — an apex inter-regulatory coordination body chaired by the Finance Minister.

Connection to this news: The draft MRMF, once finalised as a master direction, will be legally binding on all regulated entities listed in its scope — binding outcomes through RBI's regulatory authority under the Banking Regulation Act and RBI Act.

AI Ethics and Human-in-the-Loop Governance

"Human-in-the-loop" (HITL) is a design principle for AI systems that requires a human decision-maker to review, approve, or override AI-generated outputs before they take effect — or, in the case of a kill switch, to halt the system entirely. This principle is central to responsible AI frameworks globally.

India's National Strategy for Artificial Intelligence (NSAI), released by NITI Aayog in 2018, identified financial services as one of five priority sectors for AI adoption, alongside healthcare, agriculture, education, and smart cities.
The EU AI Act (2024) classifies AI in high-stakes sectors like credit scoring as "high-risk AI" requiring mandatory human oversight — a category parallel to what RBI's MRMF addresses.
The RBI framework's explainability requirement aligns with principles of algorithmic accountability: that automated decisions affecting individuals (e.g., loan rejection) must be explainable and contestable.
Credit information companies (CIBIL, Experian, Equifax India, CRIF High Mark) are explicitly included in the framework's scope, since their AI models directly determine credit access for individuals.

Connection to this news: The RBI framework is India's first sector-specific regulatory intervention requiring HITL governance for AI in finance — positioning India's central bank ahead of many global peers in formalising AI accountability standards for the financial sector.

Key Facts & Data

Framework title: Guidance on Regulatory Principles for Model Risk Management, 2026
Released by: Reserve Bank of India (PR 2026-2027/528) on June 24, 2026
Public comment deadline: July 24, 2026
Entities covered: commercial banks, small finance banks, payments banks, co-operative banks, regional rural banks, NBFCs, local area banks, AIFIs (NABARD/NHB/EXIM/SIDBI), ARCs, credit information companies
Key mandate 1: Board-approved Model Risk Management Framework (MRMF)
Key mandate 2: Kill switch — override/suspension/deactivation mechanism for AI/ML models
Key mandate 3: Explainability thresholds — AI decisions must be interpretable
Key mandate 4: No liability transfer to third-party AI vendors
Key mandate 5: Bias monitoring and adversarial testing
RBI established: 1934 (RBI Act, 1934)
Banking supervision authority: Banking Regulation Act, 1949
NBFC supervision authority: RBI Act, 1934, Chapter III-B
Related India AI policy: NITI Aayog National Strategy for AI (NSAI), 2018