It was Global Fintech Fest week — GFF 2026 in Mumbai, September 8–11, drawing over 100,000 delegates — and for once the policy calendar did not compete with the product launches; it was the main event. The RBI recognised the sector’s second self-regulatory organisation. NPCI confirmed it is building protocols — and a registry — to let AI agents pay on UPI, while its chairman drew one very bright line. The SEBI chairman ruled out delegating enforcement to a black box. The RBI Governor warned fintechs against regulatory arbitrage and told them customer data is a fiduciary duty, not a business asset. And the Prime Minister proposed a public rating system for fintech consumer protection — an idea consumer collectives have argued for for a decade. Here is the week in five stories, with the consumer’s stake in each.
1. The Second Fintech SRO: UFF Joins FACE at the Self-Regulation Table
On September 10, RBI Governor Sanjay Malhotra announced that the Reserve Bank has recognised the Unified Fintech Forum (UFF) as the second self-regulatory organisation for the fintech sector (SRO-FT) — nearly two years after the Fintech Association for Consumer Empowerment (FACE) became the first in August 2024. UFF — until April 2025 the Digital Lenders Association of India (DLAI) — had been pursuing recognition since RBI issued the SRO-FT framework in May 2024, and the regulator confirmed it examined the application against the framework’s requirements before approving. UFF’s executive committee includes UGRO Capital’s Shachindra Nath, Lendingkart’s Harshvardhan Lunia, Fi Money’s Sujith Narayanan and BharatPe’s Nalin Negi, per Economic Times. Our SROTrac tracking counts 121 UFF members against FACE’s 85 (BusinessLine reports FACE’s “members and partners” at 378 — a broader, affiliate-inclusive count).
Two SROs for one sector is a deliberate design choice, not an accident — RBI declined to grant a monopoly. But dual membership creates an obvious soft spot: a lender in both UFF and FACE now sits under two codes of conduct, and in practice the softer enforcement will set the floor. The sharper question is UFF’s DNA. It was founded as a digital-lenders’ association, and digital lending is precisely where India’s consumer harm has been most acute — forced recovery, opaque charges, contact-list harvesting, lending through white-label apps that vanish after a complaint cycle. An SRO earns its recognition not by publishing a code of conduct but by disciplining members: expelling repeat offenders, publishing grievance data, standardising recovery conduct. Neither SRO-FT has yet expelled its way into headlines.
The consumer stake: treat the SRO as a first stop, not a last resort. The RBI Banking Ombudsman (cms.rbi.org.in) remains the binding escalation path, and no SRO code can override RBI’s digital lending guidelines. What to watch: whether UFF publishes member-wise complaint statistics in its first year — that single disclosure would tell you whether self-regulation here is oversight or branding.
2. AI Agents on UPI: A Protocol, a Registry, and One Bright Line
The Reuters report of September 1 — that India is preparing a Unified Agent Protocol (UAP) letting AI agents make small UPI payments without per-transaction human approval — got its official shape this week. On September 10, ET reported that NPCI is building a registry to verify and monitor AI agents transacting on UPI, starting with UPI and possibly extending later to cards and bill payments. At GFF, NPCI’s non-executive chairman Ajay Kumar Choudhary set the philosophical boundary: an AI agent may work out what a user wants, but it should not be the thing that approves the payment — AI is adaptive and probabilistic, while payment infrastructure must stay deterministic, auditable and final, so the architecture must separate intent, authorisation and settlement. “NPCI is also examining the protocols that may be required to identify and authorise digital agents within the UPI ecosystem while preserving interoperability, auditability and settlement finality,” Choudhary said.
The reported design leans on two existing delegated-payment rails: UPI Circle, which handles who may initiate a payment, and Reserve Pay, which caps how much is at stake — banks currently block around ₹10,000 for up to 90 days, limits that may be reconsidered for agentic payments (The Daily Brief). Expected controls per Times of India: identity checks, transaction rules, spending limits, audit trails — and a liability framework whose details are not yet public. Scale explains the urgency: UPI processed 24.51 billion transactions worth ₹29.82 lakh crore in August — 791 million a day — and the private sector is already running ahead of the rulebook, with Pine Labs’ P3P protocol letting agents pay after a single upfront authorisation (Gullak’s gold-savings agent uses it), and the SBI chief floating “Know Your Agent” as the next KYC.
The consumer stake: this is the most consequential payments rulemaking since AutoPay. The bright line — agents propose, rails approve — is the right one. The unresolved half is liability: when an agent misreads intent or gets compromised, who eats the loss — the user who set the limits, the app hosting the agent, or the issuing bank? Current chargeback norms were written for humans tapping PINs. If UAP ships with “no friction” as its marketing and silence on liability as its fine print, the default will be the consumer pays. That allocation must be written into the protocol, not settled by grievance precedents years later.
3. RBI’s Posture: Same Activity Same Risk, Data as Fiduciary, No Gaming the Gaps
Across three days of GFF appearances, RBI leadership gave the most coherent statement of regulatory philosophy since the Governor took office. Malhotra’s formulation: “same activity, same risk, same regulatory treatment” regardless of who performs it — keep regulation light-touch where innovation is nascent and risk contained, and step in when activity grows to scale or consumer-conduct issues emerge. His sharpest warning was aimed at regulatory arbitrage: fintechs should not build businesses “by taking advantage of gaps between regulatory categories,” and responsibility should grow with systemic footprint — “extending beyond its own balance sheet” once payments volumes, lending books or user bases are big enough to move the system. On data, he was blunter still: fintechs must treat customer data as a fiduciary responsibility and not as business assets. The Governor also noted the RBI has engaged with more than 500 fintech entities under its innovation-friendly posture.
Deputy Governor Rohit Jain supplied the risk taxonomy: speed, concentration and opacity — none new, but each amplified by technology. His concentration warning is the one to hold: financial institutions increasingly depend on a small number of cloud, technology and AI model providers, often on overlapping datasets, so a single provider’s failure could transmit disruption across many institutions at once. Regulation, he argued, should focus on outcomes — fairness, accountability, resilience, customer protection — and “the greater the consequence of the use case, the stronger the expectations” on governance and oversight. Read between the lines and a tougher outsourcing/IT-resilience rulebook is likely coming.
The consumer stake: two of these translate into concrete protection if they harden into rules. The data-as-fiduciary framing, written into regulation, would directly constrain the alt-data habits of lending apps — transaction histories and contact graphs treated as held-in-trust rather than monetisable. And concentration risk is why your UPI dies when one bank’s switch or one cloud region hiccups; RBI supervising common dependencies, not just individual institutions, is the structural fix for the recurring “payment failed but money debited” week.
4. SEBI: “An AI-Generated Alert Is Not a Finding”
SEBI chairman Tuhin Kanta Pandey, speaking on a GFF panel moderated by NPCI’s Dilip Asbe alongside IOSCO’s deputy secretary general, drew the capital-markets version of NPCI’s bright line: enforcement and adjudication “cannot be delegated to the black box”. “An AI-generated alert is not a finding,” he said — surveillance AI may surface suspicion, but a human must investigate, establish facts and record a finding before anything touches a person’s file. His operational demand was for regulatory capacity itself: “supervisors must have people who can understand, challenge and override AI outputs.” Pandey also reported that SEBI’s Demat 2.0 pilot for corporate bonds has completed its first phase with three issuers, with secondary-market trading next in line.
The consumer stake: this cuts two ways, and both matter. On one side, AI surveillance that cannot convert suspicion directly into penalty means slower enforcement — fraud victims will not see faster restitution from this. On the other, anyone wrongly flagged — and fraud-detection models flag false positives at meaningful scale — keeps the human checkpoint that due process requires; SEBI quietly chose the accused’s rights over the regulator’s throughput, which is the correct default. Demat 2.0 is the sleeper item: India’s corporate bond market is an opaque dealer-dominated venue where retail investors get the worst pricing, and moving issuance and eventually secondary trading onto a demat-native settlement framework is the precondition for a genuine retail bond market. Watch whether the secondary-trading phase brings visible pre-trade price transparency or just electronic plumbing.
5. The PM’s Fintech Consumer-Protection Index: Rating the Rated
The Prime Minister’s September 8 keynote got its share of coverage for declaring UPI an export — pushing global UPI links to cut the 3–4% cost that 32 million NRIs pay on remittances, against a sub-1% domestic UPI cost base (press insider account). The under-covered proposal is the more interesting one for consumers: a public rating system that would score every fintech company on its consumer-protection practices — in the framing of the keynote coverage, “a credit rating for how well fintech firms treat their users.”
In principle, this institutionalises what consumer collectives do ad hoc: making conduct comparable before download. Whether it works depends on three design choices. First, who scores — if an SRO rates its own members, that is the fox auditing the henhouse; the credibility of the index requires an independent body or regulator. Second, what data — the only ground truth is outcomes: RBI Ombudsman complaint volumes and reversal rates, cyber-fraud incidence, recovery-conduct violations. A methodology built on self-reported compliance will measure marketing departments, not conduct. Third, comparability — scores must mean the same thing for a lending app and a payments app, or the index becomes a wall of misleading stars. The credit-rating precedent is the standing warning: ratings paid for by the rated get gamed.
The consumer stake: a credible index would move conduct information to the point of choice — the app-store moment where trust is actually extended. The test of seriousness arrives fast: if the first methodology draft binds scores to complaint-resolution data, take it seriously; if it launches as a badge, treat it as noise.
The week in one line: four regulators and one prime minister drew the same shape from different angles — AI may assist, the record must stay auditable, a human stays accountable, and the consumer’s data is held in trust. The unfinished sentence in every one of those drafts is liability: who pays when the machine is wrong. That is the file to watch.