As banks move from using artificial intelligence to assist employees and customers towards AI agents that can independently take actions, the biggest challenge may not be what these systems can do, but whether they can be trusted to do it consistently.State Bank of India chairman Challa Sreenivasulu Setty said financial institutions will need to build trust into agentic AI systems from the outset, arguing that the standards for AI operating in financial services will have to be significantly higher than in many other applications.“There is nothing like 99% accuracy here. You need to be 100% accurate, and every time, all the time, at population scale,” Setty said while speaking at the Global Fintech Fest 2026.He outlined what he called the three A’s of trusted AI — accuracy, accountability and access without asymmetry.Also Read: GFF 2026: Axis Bank, J.P. Morgan see banking move from AI that knows customers to AI that actsThe first is accuracy. According to Setty, average performance is not enough when AI systems are being used across a diverse customer base and potentially making or executing financial decisions. “Consistent performance at scale itself becomes a test of AI quality,” he said.The second is accountability. As AI moves beyond recommending an action to actually executing it, banks will need to be able to establish why an action was taken and who is responsible for it.“The moment AI moves from recommendation to execution, accountability becomes even more important,” Setty said, adding that systems will need an audit trail, traceability and the ability to understand why an important action was taken.This becomes particularly important as banks move towards agentic AI, where systems can act with a degree of autonomy rather than simply respond to predefined instructions.Setty said SBI is already using conventional AI and machine learning across areas such as customer service, pre-approved personal lending, credit assessment, cash-flow-based lending, fraud detection, anti-money laundering and early warning systems. The bank is now looking at the next generation of AI capabilities that can act on behalf of customers and institutions.Agents could eventually operate across the financial lifecycle, from KYC, AML and regulatory processes to fraud and mule-account detection, loan appraisal, underwriting and reconciliation, he said.But the same autonomy that makes these systems powerful also increases the potential impact of errors.“When a system can act autonomously, an error can potentially trigger a sequence of actions across multiple systems and affect customers, counterparties and institutions at machine speed,” Setty said.His third principle is access without asymmetry, which he said means AI should be available without bias and should adapt to the customer rather than requiring the customer to adapt to the technology.For India, this could become particularly important as banks attempt to take AI beyond digitally savvy customers. Setty said agentic AI could potentially democratise financial intelligence that is currently concentrated among customers who have access to wealth management and premier banking services.Also Read: GFF 2026: RBI flags speed, concentration, opacity as key risks as AI use grows in finance“The most exciting aspect of agentic AI is not automation. It is the possibility of democratising financial intelligence,” he said.The broader challenge, however, will be taking these systems to India's scale without making them prohibitively expensive. Much of the deployment could happen through mobile phones, making affordability, latency and computing efficiency important considerations, Setty said.He also argued that India's AI opportunity would depend on building models that understand the country's diversity rather than simply building larger models.Voice-first and vernacular banking could be among the key use cases, particularly for customers who are more comfortable speaking than typing or who do not use English for financial interactions.