CERT-In, the cyber wing of the IT ministry, proposed to mandate human-in-the-loop interventions in agentic AI payments above a certain threshold, according to the Digital Threat Report 2025-26.
“Mandate human-in-the-loop controls for agentic AI actions above defined financial thresholds, with full audit trails,” the report says.
In the context of AI, human-in-the-loop means that a human is monitoring or intervening in the AI workflow. In these systems, humans may provide feedback, correct errors, or make financial decisions to improve the accuracy and reliability of AI.
AI agents to make UPI payments? MeitY’s proposal comes amid reports that AI agents may soon be allowed to make UPI payments. As per Business Standard, the National Payments Corporation of India, a private non-profit company not under RTI, which runs UPI, is building a Unified Agent Protocol to enable this. However, the report did not specify how this agentic protocol would work.
Nikhil Pahwa, founder and editor of MediaNama, says NPCI must first build trust before rolling out agentic UPI payments.
“Agentic commerce fails because there is a lack of trust. People in the US who have given their cards to agents have found, at least one person claimed, that an agent paid for a $2,500 course influenced by an Instagram influencer. We need guardrails, recourse and reversibility, though that is tricky in India,” Pahwa wrote in his Reasoned newsletter.
On identity, he proposed to “give agents their own delegated handles, or to make it even simpler, delegate a handle with agent or at agent”—for example, agent-nixxin@ybl or agent@nixxin@ybl.
“Separate agent PIN, on by default. Allow users to turn it off. The moment they turn the PIN off, the limit goes down to the default level, and users have to increase it if they want,” he said.
Pahwa also proposed that users be allowed to set transaction limits in three ways:
Max amount per transaction
Max transactions and amount per month
Max transactions per day, and per week
Pahwa also said that AI agents shouldn’t have bank accounts but should instead use wallets. You can read his views on how NPCI should approach agentic payments here.
How an agentic payment protocol might work: Coinbase introduced the x402 protocol last May, which “enables AI agents to autonomously discover and procure third-party cloud resources, contextual data, and API tools—making it easier for them to achieve their targeted optimisation goals without human-in-the-loop intervention.” “This enables fully autonomous, AI-driven commerce—allowing goal-oriented agents to operate independently in an on-demand, permissionless economy.”
Cloudflare, which is allowing publishers to charge to allow agents to access content, is using payments via x402 to enable access.
Last month, Pine Labs launched an agentic payment protocol, called P3P, built on UPI, though some questions remain unanswered about data sharing. The company also open-sourced its agentic OAuth framework for authenticating agents.
Currently, agentic commerce is a tug-of-war between OpenAI’s and Google’s protocols, with contrasting views on how AI agents should handle financial transactions.
Also read: