The request
On November 17, 2026, Business Today relayed an Economic Times report that the National Payments Corporation of India (NPCI) has opened discussions with Google about Android-level safeguards against UPI fraud. MediaNama's November 19 follow-up adds detail. Fraudsters distribute malicious APKs through messaging apps, fake customer-support links and spoofed update prompts. Malware called Ghost Overlays, and variants of the GhostBat remote access trojan, then capture UPI PINs. Neither report records a response from Google, and the talks rest on one secondhand report. Treat them as a proposal, not a policy.
An unnamed payment-app official told the outlets that awareness campaigns are not enough. The industry wants to block accessibility, device-administration and display-over-other-apps permissions for apps installed outside verified stores, unless the user passes multi-step safeguards. It also wants server-side checks of device attestation keys against Google's list of revoked keys.
Why the attack is hard for apps to stop
Cyble's October 2026 analysis, reported by CXOToday, shows why payment apps feel stuck. Victims get an SMS about an unpaid e-challan and are pushed to a fake mParivahan app installed from outside the Play Store. Once the victim grants Accessibility, the malware waits for the PIN screen of Google Pay or MobiKwik. It then lays a transparent grid of 12 touch zones over the real keypad, logs each digit and replays the tap into the genuine app. A second capture method covers 60 banking, UPI, wallet and fintech apps by default.
The user is on the real app, the PIN goes to the real bank, and the transaction looks legitimate. The compromise sits in the operating system, below anything a payment app controls. This is the strongest case for OS-level action. Only the platform owner can see which permissions an unvetted app has been granted and when it is drawing over a PIN pad.
The pattern is not new. In March 2026, CloudSEK's "Digital Lutera" report described Telegram groups distributing a toolkit that hides in APKs disguised as traffic fines or wedding invitations. NPCI replied that "robust checks and safeguards are already in place." CERT-In has separately catalogued Android trojans, such as SpyMax, that spread as fake wedding-invitation APKs and ask for accessibility, notification and SMS permissions.
The case for stronger controls
Regulators have a fair point. Fraud losses fall on individual users, often people with older phones who cannot judge a permission prompt. Awareness campaigns put the burden on the victim. Google has already shown the approach works in principle. Its October 2024 announcement of an enhanced fraud protection pilot in India found that over 95 percent of installations of major fraud malware families exploiting sensitive permissions came from internet-sideloading sources. The pilot blocks such installs automatically. If the delivery channel is that concentrated, a narrow intervention there is justified.
Where proportionality matters
The risk is overreach. Android's openness is a feature of India's internet economy. Developers distribute apps to regions the Play Store serves poorly. Researchers, students and hobbyists test their own builds. Microfinance and government apps are shared directly. A blanket rule that cripples sideloaded apps would hit these users far harder than it hits organised fraud rings, which can switch to other tricks.
The proposals differ in how proportionate they are:
- Permission scoping is proportionate. Blocking accessibility, device-admin and overlay access for unverified apps, with a deliberate friction flow, targets the capability fraudsters need. A user can still install a hobby app. That app cannot read the screen or draw over a banking keypad.
- Risk-based transactions are proportionate. MediaNama reports a suggestion that devices with outdated patches or suspicious behaviour face transaction limits or step-up verification, not exclusion. Exclusion would lock out the many Indian users on older phones, who are also the ones most exposed.
- Attestation checks need care. Checking attestation keys against revoked-key lists and strengthening Play Integrity signals can help. But they hand a private platform a de facto gatekeeping role over who can use a public payment rail. The rules need to be published, appealable and applied the same way to all UPI apps.
Google has also been moving. MediaNama reports a 2026 flow for installing apps from unverified developers, with developer mode, a confirmation that the user is not being coached, a restart, a one-day wait and biometric authentication. It also reports free limited-distribution accounts for students and hobbyists. That is the right shape: friction that scales with risk, with a path left open for legitimate use.
What good policy looks like
First, measure the problem before widening the remedy. Public data on how many UPI frauds involve overlay malware, as against social engineering, would let NPCI and Google show what works. The reports cite no such figures, and without them the debate will run on anecdote.
Second, keep the obligation shared. UPI apps should not be able to push all device-level risk onto Google, and Google should not be left to design de facto payment regulation alone. A published technical standard, with an agreed baseline of signals and a clear liability allocation, serves users better than private bargaining.
Third, protect the lawful fringe. Any OS rule should have a documented route for developers and a recoverable path for users. The aim is to stop a transparent grid of touch zones sitting on a PIN pad, not to end sideloading.
The measures that work here are narrow, evidence-backed and reversible. The ones to resist are those that quietly make a permissioned platform the price of using UPI.