AI errors could pose cascading risks to financial institutions: SBI chief

SBI Chairman CS Setty says banks must strengthen AI governance through clearly defined permissions, transaction limits, audit trails and human oversight to prevent systemic risks

SBI Chairman Setty
C S Setty, Chairman, State Bank of India (SBI). (File Pic)
Subrata Panda Mumbai
4 min read Last Updated : Nov 19 2026 | 5:09 PM IST
Artificial intelligence (AI)-related errors could cascade across interconnected systems, creating significant governance risks for financial institutions. Therefore, clearly defined permissions, transaction limits, audit trails and human intervention mechanisms are very important as banks increasingly deploy autonomous systems, said C S Setty, Chairman, State Bank of India (SBI), on Thursday.
 
“While AI capabilities offer substantial productivity gains, they also create new governance risks,” Setty said while speaking at an event in Mumbai. “Institutions must clearly understand what an autonomous system is authorised to do. The information it may access, the decisions it may undertake and the circumstances in which human intervention becomes mandatory,” he said.
 
Against this backdrop, Setty highlighted some key responsibilities of boards, including ensuring that the adoption of artificial intelligence (AI) is driven by institutional purpose rather than technological momentum. Technology, he said, should address genuine business needs, whether by improving customer experience, strengthening risk management or widening access to financial services.
 
“Adoption cannot become an end in itself. This is particularly important in a country as diverse as India,” Setty said.
 
Setty cautioned that AI must not inadvertently exclude rural borrowers, small businesses or individuals without conventional credit histories, adding that financial inclusion must remain central to the design and deployment of the technology.
 
On governance frameworks, Setty said SBI was the first bank in the banking system to adopt a responsible AI policy, setting out principles, boundaries and accountability for the use of the technology across the organisation.
 
He said bank directors needed to build the capacity to assess model risks, data governance, cybersecurity and algorithmic bias, while understanding the limitations of AI.
 
“Today every bank board is expected to undergo a cybersecurity drill every year. And while it is still a learning process, I think now we have to integrate AI learning,” he said, adding that directors needed to understand the risks associated with AI rather than necessarily learn how to implement it. 
He also referred to the Reserve Bank of India’s AI policy, which, he said, places ultimate responsibility for AI governance on the board. At the same time, AI could improve board effectiveness by highlighting emerging risks, simplifying complex information and supporting better-informed decisions, allowing directors to devote greater attention to strategy and long-term priorities.
 
Emphasising the need to retain human accountability, Setty said AI could recommend or undertake actions, but ultimate responsibility for institutional decisions must remain with the board. “If an institution cannot adequately explain why an AI-backed process denied a loan, flagged a transaction or initiated an action, it faces serious governance challenges,” he said, underlining the importance of explainability, traceability and independent validation.
 
He added that AI was most effective in situations where parameters could be defined, relevant data was available and outcomes could be assessed. However, during crises, unexpected changes in the operating environment or strategic decisions involving unfamiliar circumstances, the collective experience and independent judgement of a board remained indispensable.
 
“The pace of AI adoption must remain commensurate with an institution’s capacity to govern the technology, particularly where decisions carry significant financial, legal, ethical or reputational consequences,” he said.
 
Setty said corporate governance was under particular stress when the pace of technological change exceeded the frameworks designed to govern it. Boards must therefore ensure that innovation advances without compromising institutional integrity and public trust.
 
“AI should complement, not substitute, the judgement and oversight of boards, strengthening their ability to discharge their responsibility while leaving accountability firmly where it belongs,” he said.
 
He highlighted that the country’s largest lender had been using artificial intelligence and machine learning, particularly in fraud risk management and business operations.
 
Among its initiatives is a patented AI and machine learning model that identifies branches that are outliers from a fraud-management perspective. The model evaluates around 23,000 branches, classifying them into very high-risk, high-risk, medium-risk and low-risk categories. The assessment helps the bank prioritise branch audits, he said.
 
However, Setty said the focus was now shifting from individual AI use cases to enterprise-level adoption, where governance would become particularly important for large institutions.
 
He identified fairness, reliability, explainability, security and accountability as critical elements of AI governance. Model bias, he said, was not merely a theoretical concern for a bank such as SBI, which serves a highly diversified customer base. “Fairness and preventing model bias are extremely important from the governance point of view,” he said.

More From This Section

Topics :artifical intelligencesbicybersecurity

First Published: Nov 19 2026 | 5:09 PM IST

Next Story