CERT-In floats mandatory human-in-the-loop for high-value agentic AI payments in India

In plain terms

India's cyber-security agency CERT-In (under MeitY) has proposed rules that would require a human intervention step in AI-driven payment flows above a "high-value" threshold. If enacted, fully autonomous agentic checkouts couldn't run untouched on Indian consumer rails past that ceiling - a first-mover position among major economies.

Context

Agentic payments - where an AI agent completes a purchase, transfer or subscription without a per-action user confirmation - are being pushed by OpenAI (ChatGPT Agent), Anthropic (Claude with Computer Use) and payment providers (Stripe, Visa's Intelligent Commerce). India's UPI is the dominant national payment rail; the RBI has historically been strict on delegated-payment models. CERT-In's draft sits alongside the wider IndiaAI framework and follows the earlier ministry-level pause on OpenAI / Anthropic deployments. NPCI and Indian fintechs are separately developing protocols to let AI agents transact over UPI.

The data

Per Medianama (Jul 16, 2026), CERT-In has proposed mandatory human oversight specifically for high-value agentic AI payments. The exact "high-value" threshold, consultation window and enforcement mechanism are not yet public. No RBI joint communication has been published at the time of the article. NPCI's agentic-UPI protocols are still in industry consultation.

Analysis

Two forces meet. MeitY's ongoing push for domestic AI sovereignty (Bhashini, IndiaAI, ministry-level pauses on foreign APIs) intersects with the RBI's structural conservatism on payments. A mandatory human step at high value doesn't kill agentic commerce - small purchases can flow autonomously - but it kneecaps the marquee use cases Western agent providers are pricing into their roadmaps: travel booking, B2B procurement, subscription bundles. It fits a broader pattern: India narrowing autonomy scope where foreign-provided AI intersects with critical infrastructure.

Scenarios

Base case: rule enacted with the high-value threshold set at consumer-meaningful levels (₹5,000 - ₹25,000 range often used elsewhere for step-up authentication), turning agentic checkout above that into a two-step flow. Local upside: domestic AI stack (Sarvam, Krutrim) certified faster than foreign providers adapt. Bear: industry pressure stalls consultation, rule reappears in weaker form as a self-attested compliance box.

So what

For readers building agentic commerce: India - one of the world's most active payment ecosystems by transaction count - may become the first big jurisdiction where "no human step" isn't a design option above a threshold. Design the two-step flow now, or lose the India user for anything but low-value purchases.