Key Details
In his keynote address, A Vision for Responsible AI, Resilient Banking, RBI Deputy Governor Shirish Chandra Murmu presented five principles for deploying AI without weakening resilience, fairness or customer protection. The speech signals supervisory expectations rather than introducing new regulations.
Principle | What Banks Are Expected to Do |
|---|---|
Include | Use AI and alternative data to reach people and businesses that conventional credit systems cannot adequately assess |
Adapt | Upgrade technology, processes and employee skills as banking models and customer expectations change |
Understand | Maintain an inventory of AI systems—including models embedded in vendor products—and understand their limitations |
Safeguard | Test systems before deployment and periodically thereafter, while preserving alternatives for critical operations |
Explain | Make consequential decisions understandable and contestable, with an authorised person able to intervene |
AI Should Expand Credit Access, Not Only Cut Costs
India’s banking system enters the AI transition from a position of financial strength: its capital adequacy ratio stands at 17.7%, gross non-performing assets have fallen to 1.8%, and annual profit after tax exceeds ₹4 lakh crore.
Yet credit growth does not necessarily mean wider access. The share of new businesses entering the formal credit system declined from 52% in 2022–23 to 42% in 2025–26, even as outstanding commercial credit grew by 14%.
The Deputy Governor urged banks to use AI to understand borrowers who lack conventional collateral, financial statements or credit histories (Credit Invisibles). With consent and appropriate safeguards, information such as cash flows, GST filings, utility payments, agricultural data and seasonal income patterns could help assess such borrowers more accurately.
What Are the “Credit Invisibles”? These are individuals or businesses whose repayment capacity is difficult to establish through conventional records. They may have viable incomes but lack a formal credit history, documented financial statements or acceptable collateral.
Banks Remain Responsible for Automated Decisions
The speech draws a clear line between using AI and transferring accountability to it. When an automated system declines a loan, reduces a credit limit, restricts an account or denies a claim:
The customer should know that an automated system is involved.
The basis of the decision should be understandable and communicable.
The customer should have access to a human reviewer with authority to reverse an incorrect decision.
The governing principle is direct: a machine may reach a decision, but a person must own it. Banks therefore remain accountable even when models are supplied by fintech partners or embedded in third-party products.
This approach aligns with the SEBI Chairman’s address delivered on the same day. Both regulators place responsibility for AI outcomes on the regulated institution, including when a model is supplied by a third party. SEBI’s proposed AI/ML guidelines are expected to go further by requiring human-in-the-loop controls and a “kill switch”, indicating growing convergence around human oversight across India’s financial regulators.
Shared Technology Can Create System-Wide Risks
AI can help banks connect complaints, audit findings, transaction patterns and system logs to identify fraud, misconduct, cyber threats or operational weaknesses earlier. RBI initiatives such as MuleHunter.ai and the Digital Payments Intelligence Platform illustrate this collective use of intelligence.
The same interconnectedness also creates concentration risk. If several banks depend on identical models, data sources, cloud infrastructure or technology providers, a single error or outage could affect multiple institutions simultaneously. Banks are therefore expected to maintain credible alternatives and avoid excessive dependence on common providers.
Policy Relevance
For Indian banking policy, the address shifts the AI debate from adoption rates to who benefits, who remains accountable and how failures are contained.
The immediate regulatory challenge is to translate these principles into verifiable practice: model inventories should cover third-party systems; testing should examine outcomes for vulnerable customer groups; automated decisions should offer meaningful explanations; and escalation routes should lead to people empowered to correct errors.
The speech also offers a more demanding measure of AI productivity. Success should not be judged only through lower costs or faster processing, but through new borrowers reached, grievances resolved, risks identified earlier and customer decisions made fairer.
Relevant Question for Policy Stakeholders: How should RBI ensure that banks can demonstrate and not merely assert that their AI systems expand access, treat customers fairly and remain subject to effective human control?
Follow the Full Speech Here: A Vision for Responsible AI, Resilient Banking