NPCI explores agentic AI layer for faster UPI compliance cycles

As updates to India’s real-time payments system, the Unified Payments Interface (UPI), come thick and fast, the National Payments Corporation of India (NPCI) is exploring agent-to-agent (A2A) workflows to speed up compliance cycles, sources said.

 

A2A flows could offer an early glimpse of artificial intelligence (AI) agents at the apex retail payments body interacting with similar agents at India’s leading banks to hasten compliance as the UPI system undergoes successive iterations.

 

Currently, banks require four to eight weeks for certification with UPI’s operating circulars (OCs) that NPCI releases from time to time.

 

The organisation is envisioning an environment where this could be cut to a week or 10 days, said a person with knowledge of the matter. “A product like UPI, which is so dynamic, requires some automation. This is a clear use case where automation can be applied.”

 

While development remains in its early stages, NPCI is expected to reach out to banks and technology service providers to build the system, and tangible benefits could be more than a year away, said a second person.

 

“Agents would look out continuously to solve a problem. Once an end-point is defined, agents are capable of working towards it. The main purpose of certification is to ensure that when banks and NPCI are communicating with one another, all the attributes are operating accurately,” said one of the persons quoted above.

 

An email sent to NPCI on the matter did not elicit a response until press time.

 

OCs related to UPI are policy and technical directives issued by NPCI to ecosystem participants including banks and fintech companies. These circulars are issued regularly and are aimed at ensuring uniform upkeep of the payments system that spans multiple banks and third-party application providers (TPAPs) such as PhonePe, Google Pay, Paytm and others. 

 

The focus on fast-tracking compliance for evolving systems like UPI comes at a time when the system undergoes multiple iterations every year along with the launch of new use cases.

 

For instance, in 2025–26, NPCI issued more than 30 UPI-related OCs for compliance across system participants. An A2A workflow is expected to ease technical and compliance bandwidth for them.

 

Building agents

 

Even as the payments body pushes for efficiency gains through AI agents, it will not deploy its agents within the environment of a bank.

 

Banks will need to build their own agents. In return, agents on both sides would communicate over secure protocols.

 

That said, a human-in-the-loop would continue for final authorisation, testing and certification. “At this stage, it is an effort. Most interactions would move to agentic flows in the future. Both agents work in different contexts and environments and should be able to communicate with each other,” said a source.

 

Another person said: “NPCI has a list of different dependencies when it comes to certification of OCs. With A2A communication, these things can be addressed.”

 

A2A protocols

 

Google’s A2A is an open protocol that provides a standard way for agents to collaborate with each other, regardless of the underlying framework or vendor, according to the company’s blog. It is capable of providing real-time feedback, notifications, and state updates to its users.

 

NPCI may explore such open protocols or develop custom ones.

 

Google says A2A facilitates interaction between two agents -- ‘client’ and ‘remote’. The client agent formulates and communicates key tasks, while remote agents act on these.

 

NPCI has set the ball rolling for other AI initiatives in the past.

 

Earlier this year, it rolled out FiMI (Finance Model for India), a domain-specific language model for the Indian payments ecosystem.  FiMI is built to natively understand the complexity of Indian payment systems including UPI, along with transaction dispute handling, management of payment mandates and other functions 

Agents at work? 

• NPCI is looking at agent-to-agent workflows to automate UPI compliance processes
• Banks currently take four to eight weeks for certification with UPI’s operating circulars
• Agentic systems could reduce certification timelines to 10 days
• Banks must build in-house AI agents communicating via secure protocols
• Human oversight to be retained for approval, testing, certification processes