India plans to let AI agents pay on UPI without each approval

India is preparing to let AI agents spend real money on its Unified Payments Interface without asking the user to approve each transaction, according to reporting based on people familiar with the plans. UPI is one of the largest retail payment networks in the world, which would make it among the first at that scale to carry autonomous agent payments.

The mechanism described is delegation rather than a blank cheque. Users would set rule-based instructions in advance — how much an agent may spend, on what, and within which limits — and the National Payments Corporation of India is expected to provide infrastructure that merchants can integrate with directly. Spending caps, audit trails and identity checks are described as built in.

The framework has been reported under the name Unified Agent Protocol and is expected to be introduced at the Global Fintech Fest in Mumbai.

The risk side is being discussed alongside it. If the safeguards are vague, fraud controls weak, or liability unclear when an agent buys something the user did not want, the same design could expose a very large number of people to a new category of financial loss. The Reserve Bank of India and the market regulator SEBI have separately been considering “kill switch” mechanisms that would let a user halt transactions from their account during a suspected fraud — a control that becomes more consequential once software can spend on its own.

The details here come from reporting rather than a published regulation, and the final rules may differ from what has been described.