All articles Fraud & ATO

SMS Pumping Fraud: Detection and Prevention

Glass message capsules rising from a dark grid, one held back by a glowing blue ring

Every one-time-password endpoint has a meter running on it. SMS pumping fraud — also called SMS toll fraud — is abuse of that meter: someone drives traffic through your OTP form so your account sends far more messages than your users asked for, and the profit splits between whoever controls the destination numbers and the operators behind them. Nothing is stolen from your database. You just pay for thousands of codes nobody ordered.

The pattern is deliberately boring: no exploit to patch, no credential stuffing. Your own form accepts a number, calls your provider and sends a code. The abuse lives in the volume and in where the numbers point.

Below is how the abuse works at the request level, why per-IP limits miss it, and how device, number and network signals become the caps that stop it.

How toll fraud works on an OTP endpoint

Number ranges belong to operators. Some take part in revenue-share arrangements: a message terminated to their range is billed by your messaging provider at a premium, and the operator passes a share to whoever generated the traffic. An attacker needs two things — a list of numbers on such a range and a public endpoint that triggers a send. Your signup form, password reset or “resend code” button is enough.

You will usually see one of three shapes:

  • Bulk single-shot. Thousands of numbers, each used once, inside a narrow window.
  • Slow drip. A few numbers per hour from many addresses, under your alert threshold.
  • Resend loops. A small set of numbers hammering the resend route, which usually has weaker limits than the first send.

The first is obvious after the invoice lands. The second costs you quietly for months.

Why per-IP limits don’t stop it

A per-IP limit assumes one sender is one machine. Pumping traffic arrives through proxies, mobile carriers and rented device farms, and the address rotates faster than your counter window. Tighten the limit and the first people you hurt are real users behind a corporate NAT or a shared mobile gateway.

One glowing node sending thin blue lines to many distant nodes on a dark surface

IP is also the wrong dimension. The attacker needs a destination number, a device that fills your form and a network that reaches you — count those. Our bot detection guide covers the automation signals in depth.

The four counters that stop pumping

Per device

Each visitor gets a stable identifier, so a script that reloads, clears storage or opens a new tab still lands on the same counter. Device signals also expose the tampering that travels with pumping: automation flags, headless browsers, mismatched timezone and language.

Per number and prefix

Count per normalized number, and per country and prefix. A prefix you never legitimately serve — or one that suddenly dominates your sends — needs its own budget instead of a share of the global one.

Per network

Datacenter ASNs, residential proxies, Tor exits and VPNs are signals, not verdicts. A datacenter address that fills a phone field in under a second is a different case from a corporate VPN at lunchtime.

The block list

Numbers and prefixes you never want to message, kept apart from the counters so they apply before anything is sent.

From counters to a decision

A cap should decide, not slow everyone down. Each check returns one of three outcomes with a reason code:

Four frosted glass panels with abstract fingerprint, number, network and bar shapes lit in blue

  • Allow. Budgets have room; your server sends the code.
  • Step-up. Ask for something the script lacks — a captcha, email verification, a card or passkey — before the send.
  • Block. Refuse the request and drop it from the queue.

Concretely: codes per device per hour, per number per day, per prefix per hour, plus the block list. When a budget is spent the answer is step-up or block, never a longer queue that keeps spending.

Prynt Control never blocks on its own. It returns the decision, the reason and a suggested step; your server chooses to send, challenge or refuse. Pseudo-code:

decision = check("sms", visitorId, phone, ip)
if decision.action == "block": return 429
send_code(phone)

The same decision covers the rest of signup, so form-level rules alone leave gaps — see form protection.

Watch it before you enforce

Control starts with a 14-day mirror period: every decision is recorded, nothing is blocked. Compare those decisions with your real send volume — which prefixes dominate, how many requests would have been stepped up, whether your thresholds sit above normal traffic. Then enforce one cap at a time so you can attribute the change.

Keep numbers as hashes, not records

Per-number caps need a counter key, not a phone book. Control keeps phone numbers and cards only as keyed hashes, so caps keep working while your database stops holding a directory of personal data. If you hash yourself, use a keyed function and a rotating key — a bare hash of a phone number can be brute-forced over a small keyspace. This is an engineering decision, not legal advice; confirm retention with your own counsel.

A black key on dark glass whose pattern dissolves into fine blue pixels

FAQ

Is SMS pumping fraud the same as SMS toll fraud?

It is the OTP-specific member of the toll fraud family, alongside international revenue share and premium-rate abuse. Most SaaS teams meet it at signup.

Isn’t a CAPTCHA enough?

It stops naive scripts and adds friction for everyone else, but it does not limit how many codes one number or prefix can burn. Use it as one step-up inside a capped flow.

Do we need raw phone numbers to count per number?

No. A keyed hash is enough to count and compare. Keep the raw number only as long as the send requires.

OTP abuse is a budget problem wearing a security costume. Decide what a device, a number and a network may spend per hour, watch the decisions in mirror mode, then enforce. Prynt Control handles the sms use case next to signup, login and checkout, so one policy covers the whole flow — see how it fits on the Prynt platform.

Try it free

Prynt is device intelligence with a free plan: 5,000 identifications a month, visitor IDs with confidence and bot detection — no credit card. Upgrade to Pro for VPN, proxy, tampering and behavioral signals.

Keep reading