SBI calls for ‘know your agent’ rules as AI moves into banking transactions
SBI Chairman CS Setty has urged a “know your agent” framework covering AI-agent identity, customer consent, transaction limits and audit trails. The bank is expanding AI use in lending, fraud detection and service while pushing for affordable India-focused AI infrastructure.
What happened
State Bank of India · SBI Chairman CS Setty called for a “know your agent” framework as AI agents enter banking transactions, covering identity, consent, limits
Key facts
- Three principles for agentic AI trust: accuracy, accountability and access without asymmetry
Why this matters
The emerging need for AI-agent identity, consent and transaction-control layers makes governance, fraud, workflow-audit and India-focused AI infrastructure targets strategically relevant.
What to watch
- RBI consultation, guidance or sandbox language covering AI-agent authentication, liability, consent and transaction authorization.
- SBI pilot announcements specifying whether agents can execute payments, modify account details, approve loans or only recommend actions.
- UPI, NPCI, Account Aggregator or digital-identity standards adding machine-agent credentials or delegated mandates.
- Procurement and partnership announcements involving AI governance, fraud, identity verification, auditability and agent-observability vendors.
- Reported incidents involving AI-driven mis-selling, unauthorized transfers, fraud losses or disputed customer consent.
- SBI and large private banks will formalize agent registries, delegated-permission models, transaction thresholds and immutable audit-log standards.
- Payment networks, UPI ecosystem participants and banking technology vendors will position identity, consent-management, monitoring and explainability tools as core agentic-AI infrastructure.
- Banks will prioritize contained deployments in fraud operations, loan servicing, collections and employee copilots before enabling agents to initiate customer transactions.
- Retailers using conversational commerce or embedded finance will need to distinguish AI-assisted checkout from autonomous agent-initiated payments and upgrade consent capture accordingly.