Will AI be making UPI payments for you now? Here's how it may work

Imagine asking your AI assistant to book a Mumbai-Delhi flight for next Tuesday, and it doesn't just show you options—it picks the best option on its own, fills in your details, and pays for it, all without you doing anything beyond giving the prompt.

 

Today, artificial intelligence (AI) agents can already search, compare and recommend options for tasks such as booking flights or shopping online. But the process comes to a halt when it is time to make the payment through the Unified Payments Interface (UPI). This is because only users and authorised payment apps, and not AI agents, can initiate UPI transactions.

 

 

Why can't AI simply pay today?

 

UPI supports features such as AutoPay and Reserve Pay, which allow users to authorise recurring or future payments by setting a spending limit and authenticating it once with their UPI PIN. These are commonly used for subscription services such as Netflix or other pre-authorised transactions.

 

But that is not the same as allowing AI agents to make independent purchasing decisions and execute payments on a user's behalf. To bridge this gap, NPCI is reportedly developing a Unified Agent Protocol (UAP), a proposed framework to verify and authorise AI agents within the UPI ecosystem.

 

What could UAP change?

 

The UAP is being designed to create a trusted, common, interoperable infrast­ructure through which AI agents can be registered, verified, and authorised to transact across the UPI ecosystem without changing the underlying rails of the payments system, Business Standard reported.

 

UAP will not replace UPI. Instead, it will act as a verification layer on top of the existing system, which will allow trusted AI agents to securely interact with UPI while preserving its existing payment infrastructure.

 

If implemented, UAP would give the payments ecosystem a way to verify that an AI agent is acting with the user's consent before a transaction is processed.

 

How might it work?

 

While there are few details available about how it will function, UAP is expected to sit on top of the existing UPI infrastructure. The payment flow could look something like this:

 

User gives an instruction → The AI assistant compares options, selects the best one and generates a payment request → UAP verifies that the AI agent is registered, trusted and authorised to act on the user's behalf → The request is routed through UPI → The user either approves the transaction, or, for pre-approved payments within set limits, the payment is completed automatically, depending on the applicable rules.

 

Will AI get full access to your money?

 

No, not necessarily. While the final framework is yet to be announced, the system is expected to work much like existing UPI features such as AutoPay and Reserve Pay, where users pre-authorise a spending limit instead of giving unrestricted access to their bank account.

 

The proposed protocol is expected to verify whether an AI agent is authorised to act on a user's behalf, define the limits of that authority, and establish accountability if those limits are exceeded, Business Standard reported earlier. The final safeguards, however, will depend on how NPCI and the Reserve Bank of India (RBI) design the framework.

 

The NPCI will also require the RBI's approval before rolling out the UAP.

 

What can go wrong?

 

As payment systems evolve to accommodate AI agents, so will cybercriminals and fraudsters look to exploit them. Allowing AI assistants to make purchases on a user's behalf introduces a new set of security and consumer protection challenges.

 

While the technology is still nascent, it introduces new risks, including fake AI agents, unauthorised or mistaken payments, overbroad user permissions and unresolved questions over liability if something goes wrong.

 

Why does it matter?

 

If implemented, UAP could place India among the first countries to build national infrastructure for agentic payments. It could transform UPI from a payment system that users operate manually into one that trusted AI agents can use on their behalf.