Who Pays when the Machine Pays? Legal Liability for Agentic UPI Transactions in India

India’s Unified Payments Interface (UPI) is entering a potentially significant transition from human-initiated digital payments to payments executed by software agents acting under pre-authorised instructions. In September 2026, reporting indicated that the National Payments Corporation of India (NPCI) was preparing a Unified Agent Protocol under which artificial-intelligence agents could execute limited UPI payments without requiring transaction-by-transaction approval. This development raises a legal question that existing payment regulation was not designed to answer: when a customer has authorised an AI agent but the agent subsequently makes an erroneous, excessive, manipulated or otherwise disputed payment, who bears the loss? This article examines whether India’s existing framework—including the Payment and Settlement Systems Act, 2007, Reserve Bank of India directions on unauthorised electronic transactions, UPI mandate architecture and consumer-protection law—adequately allocates responsibility for agentic transactions. It argues that agentic payment disputes cannot simply be treated as either ordinary authorised transactions or conventional unauthorised fraud. They occupy an intermediate category in which human authority is delegated but machine judgement determines whether, when and to whom a payment is made. Using doctrinal legal analysis, regulatory analysis, recent empirical evidence on UPI scale and comparative consideration of emerging agentic-payment safeguards, the article proposes a layered liability model based on control, authentication, scope of delegation, auditability and system failure. It recommends revocable agent credentials, transaction and temporal limits, machine-readable mandates, mandatory audit trails, rapid suspension, consumer-first provisional redress and allocation of responsibility among banks, payment intermediaries, merchants and AI providers. The objective is to preserve the efficiency of agentic commerce while ensuring that automation does not dilute consumer protection.