Exploring AI-Agent Spending Capabilities in UPI System | The Calibre Brief Ep. 12
About this episode
This episode of The Calibre Brief examines India's UPI system and its potential evolution toward allowing AI agents to make payments. Our investigation confirms that NPCI has explored agent-initiated payments but has not established a public framework for such transactions.
UPI currently enables delegated payments through mechanisms like UPI Circle and UPI Autopay. These tools allow limited delegation with caps and revocation. The challenge remains in verifying agent identity and determining liability when an autonomous transaction goes wrong.
The episode discusses unresolved issues, including agent identification, transaction limits, and cross-border regulation, emphasizing that no public system for agent payments is currently live in India. This evolving conversation highlights the complexity of moving from human to machine payment systems without compromising safety or accountability.
Key topics
Key moments
- 0:07 India has built a payment network that moves money at extraordinary…
- 0:20 Picture a UPI Circle spending envelope: one account holder lets…
- 0:48 UPI already lets users prearrange some payments through Autopay…
- 1:15 That design exposes the complication
- 1:49 That gap matters on a rail handling hundreds of millions of payments…
- 2:10 UPI can prove money moved
- 2:21 This is The Calibre Brief, from Calibre Code USA
Transcript
Show transcript
Welcome to The Calibre Brief, a Calibre Code USA production.
India has built a payment network that moves money at extraordinary scale. The harder question is whether it can safely trust software with the same privilege.
Picture a UPI Circle spending envelope: one account holder lets another person spend independently, yet keeps a 15,000-rupee monthly cap and an app-based off switch. UPI handled more than 600 million transactions a day in 2025. Reported plans ask what happens when the delegated spender is software rather than a person the holder knows.
UPI already lets users prearrange some payments through Autopay, using consent, schedules, limits, and revocation instead of a new approval each time. UPI Circle extends that logic to another human. Reporting on agentic payments contemplated a separate digital identity for each agent, verifiable and revocable, with limits attached to the agent itself.
That design exposes the complication. A payment can be authorized by setup and still violate the user's instructions in context. Current rules address unauthorized electronic transactions, and disputes can cover failures, duplicate transfers, or money sent astray. They do not clearly answer for software that misreads an instruction, is manipulated, or spends rapidly within delegated authority. No public framework has assigned responsibility for that case.
That gap matters on a rail handling hundreds of millions of payments each day. The system would need to show which agent acted, what limits governed it, and how a user could stop it without closing the whole account. It would also need a clear path when delegation becomes a loss.
UPI can prove money moved. The missing piece is proof of authority: what the agent was permitted to do, and for whom.
This is The Calibre Brief, from Calibre Code USA. If an agent stays within a cap but defies instructions, who should pay? Comment, like and subscribe on YouTube, and follow the podcast.
