Fraud

When the Victim Presses Send: Payment Fraud in 2026

November 11, 20265 min read

The fastest-growing fraud doesn't break in. It persuades someone to make the payment themselves. Here is what this year's numbers from India and abroad say, why the usual controls miss it, and what detection can still catch.

Most people picture fraud as a break-in: a stolen card, a guessed password, a cloned SIM. The kind that is growing now doesn't break anything. The account holder logs in on their own phone, passes every check and sends the money, because someone convinced them to.

I investigate suspicious transactions and accounts for a living, and I'm building a real-time payment fraud detection engine. This post is my read of what the public numbers say about 2026, and what they mean for anyone designing controls.

The numbers

Start with India. The Reserve Bank of India's Payment System Indicators for September 2026 record 3.58 lakh domestic payment frauds worth ₹493 crore in that one month. A year earlier, September 2025, the figures were 3.03 lakh and ₹431 crore. That is a rise of about 18% by count and 14% by value. June, July and August 2026 were all above 3.4 lakh too.

Two details matter when you read that table. The data comes from the Central Payments Fraud Information Registry, where banks, wallet issuers and card issuers report cases. And the RBI marks it as provisional. It counts what was reported as fraud, so it is a floor, not a ceiling.

Complaints tell a bigger story. In a Lok Sabha answer on 21 July 2026, the Ministry of Home Affairs said that from 2021 to 2025 the national cyber crime portal received more than 65.89 lakh financial fraud complaints, with more than ₹55,050 crore reported lost. Against that, the 1930 helpline system had saved more than ₹11,158 crore by the end of June 2026.

It isn't only India, and it isn't only consumers. In the Association for Financial Professionals' 2026 survey of 465 corporate finance professionals, 76% of organisations said they experienced attempted or actual payments fraud in 2025. AFP says business email compromise is still the most common form, affecting roughly three-quarters of organisations.

Authorised, and still fraud

The name for this is authorised push payment fraud, or APP fraud. In April 2026 the RBI published a discussion paper called "Exploring safeguards in digital payments to curb frauds". The Finance Ministry described it to the Lok Sabha in one line that is worth quoting in full.

“…highlighting the rapid growth of digital payments in India and the corresponding rise in Authorized Push Payment (APP) frauds, driven largely by social engineering and instant fund transfers.”

— Ministry of Finance, Lok Sabha Unstarred Question 2432, 3 August 2026

Those two ingredients explain why it is hard to stop. Social engineering means the legitimate user performs the transaction, so the password, the device and the location all look right. Instant transfers mean the money has moved on before anyone can ask a question.

The scripts change every few months, but they fall into a handful of families:

  • Borrowed authority. A caller poses as the police, customs or a bank's fraud team and keeps the victim on a call until they move money "for verification".
  • Investment and trading groups that show fake profits on a fake dashboard, until the withdrawal never arrives.
  • Fake customer support and KYC updates that end with a screen-sharing app installed on the victim's phone.
  • Collect requests and QR codes presented as a way to receive money, when approving them sends it.

Deepfakes make an old trick cheaper

Impersonation used to need a convincing actor. Now it needs a voice sample and a photo. AFP's 2026 outlook describes a case in which "a deepfake video call impersonating a CFO requested an urgent payment via WhatsApp". In AFP's words, the fraudster "may have succeeded if not for the company's 'stop, think and ask' policy that required additional validation."

I find that detail more useful than any statistic. What protected the payment was not a tool that spotted the deepfake. It was a rule: an urgent payment request gets verified on a second channel, however real the person on the screen looks. AFP's survey looked at AI-enabled fraud and deepfakes for the first time this year, which tells you how new the measurement is.

Where the money goes

Stolen money has to land somewhere, and it usually lands in a mule account: one opened, rented or taken over to receive funds and pass them on within minutes. For a detection team this is the encouraging part. The sending side of an APP fraud looks legitimate by design. The receiving side often does not.

What detection can still see

If identity checks pass, behaviour is what's left. These are the signals I would build first:

  • A first payment to a new payee that is also large for this customer. An amount far outside a customer's own history says more than any fixed limit.
  • A payee added and paid in the same session, especially straight after a new device or a password reset.
  • Velocity on the receiving side: an account that collects from many unrelated senders and empties itself quickly.
  • Small probes before a large transfer. Card testing has the same shape: several cards added, then ₹1 charges to see which ones work.
  • Order of events. The sequence matters more than any single event, which is why my fraud engine keeps one ordered stream of events instead of separate feeds.

To be clear about where my own project stands: it is at the first of seven milestones. It ingests and stores payment events today, and detection starts with the next milestone. That next layer is built around exactly these signals: velocity counters, amount z-scores and impossible travel.

Friction is a control

Product teams dislike friction, and for everyday payments they are mostly right. For a first-time, high-value transfer to a new payee, a short pause and a plain-language warning is a control, not a bug. Regulators are moving the same way: the Finance Ministry's description of the RBI paper says its measures "aim at safeguarding vulnerable customer segments undertaking specified digital payment transactions."

The cheapest control is still the one in the AFP story. Stop, think and ask. Verify on a second channel. A real CFO, bank or police officer can wait ten minutes.

If you remember three things

  • The fraud that is growing passes authentication, so look at behaviour and at the receiving account.
  • Speed helps the fraudster. A short delay on an unusual first-time transfer buys the time an investigation needs.
  • A flag nobody can explain is just noise. Whatever you detect, be able to say why.

sources

  1. Payment System Indicators, September 2026 (Part V: Domestic Payment Frauds) (opens in a new tab)Reserve Bank of India · provisional data
  2. Lok Sabha Unstarred Question No. 251: National Cyber Crime Data (opens in a new tab)Ministry of Home Affairs · answered 21 July 2026
  3. Government shares digital payment fraud data and RBI anti-fraud measures (Lok Sabha Unstarred Question No. 2432) (opens in a new tab)TaxGuru · 4 August 2026
  4. 2026 AFP Payments Fraud and Control Survey Report (opens in a new tab)Association for Financial Professionals
  5. The 2026 Payments Fraud Outlook: Lessons from Real-World Fraud Scenarios (opens in a new tab)Association for Financial Professionals · 1 May 2026