Skip to content
Back to Blog
Toll Fraud Prevention 2026: Protect Your AI Telephony
SecurityFraudTelephonyNovember 4, 202612 min

Toll Fraud Prevention 2026: Protect Your AI Telephony

Toll fraud is billing fraud in telephony: criminals make your line, your phone system or your AI phone assistant place expensive calls – to premium-rate numbers or international destinations where they earn a share of the charges. You get the bill. Effective toll fraud prevention rests on three levers: block destination countries and special-rate numbers, check every number before dialing (including transfer numbers), and cap costs technically, for example with prepaid credit instead of an open invoice.

This article focuses on cost and billing fraud. Manipulated conversations, voice cloning and prompt injection are covered in AI Telephony Security – Protection Against Voice Bot Fraud. Both belong together, but the safeguards are different.

What is toll fraud?

Toll fraud is any type of fraud in which someone triggers paid phone calls or text messages through another party's account without authorization in order to earn a share of the charges. The account holder pays; the fraudster collects through revenue sharing on expensive destination numbers.

The problem is old, but it has not shrunk. The industry association CFCA (Communications Fraud Control Association) estimated global telecom fraud losses at around 38.95 billion US dollars in its 2023 Global Fraud Loss Survey (summarized by TNS). In its developer documentation, Twilio estimates losses from International Revenue Share Fraud alone at around 10 billion US dollars per year.

Number misuse is an everyday problem in Germany, too. According to a press release dated 9 February 2026, Germany's Federal Network Agency (Bundesnetzagentur) received 85,158 complaints in 2025, switched off around 6,200 numbers and banned billing and debt collection for more than 2,000 numbers. Topics included router hacking, expensive hold queues and fake hotlines.

Which scams are behind toll fraud?

The three main scams are IRSF (calls to expensive international numbers with revenue sharing), Wangiri (one-ring calls that lure you into an expensive callback) and SMS pumping (mass-triggered text messages to numbers where fraudsters earn a cut). All three work on the same principle: every minute or message makes the fraudster money.

IRSF – International Revenue Share Fraud

Fraudsters lease number blocks in countries with high termination rates or from providers of international premium-rate numbers. Then they drive as many long calls there as possible – through hacked phone systems, compromised cloud accounts or automated systems that dial on request. The charges are split between carriers and the number holder. Twilio states that customers are responsible for all usage on their accounts.

Wangiri – one-ring calls

"Wangiri" is Japanese and roughly means "one ring and hang up". The phone rings briefly and an unknown number stays on the display. Whoever calls back reaches an expensive destination and is kept on the line with recorded announcements. On its page about ping calls, the Bundesnetzagentur warns about country codes that look like German area codes – for example Madagascar (00261) and Koblenz (0261).

SMS pumping (artificially inflated traffic)

Here, bots attack web forms that send an SMS code – sign-up, login, demo request. They enter thousands of numbers from expensive ranges, every message costs you money, and the fraudster takes a share. Twilio's guide to SMS pumping describes it as fake traffic generated to inflate costs.

Why is AI telephony a new entry point for toll fraud?

AI telephony is exposed to toll fraud because a voice agent can dial at the push of a button, around the clock and without human oversight – any number that a form, an API or a CRM list supplies. A person would hesitate if asked to call a number in East Africa 40 times in a row. An unprotected system would not.

Typical attack points in AI telephony:

  • Callback forms and demo calls: "Enter your number and our AI will call you" is a strong sales pitch – and an invitation for bots to enter expensive destination numbers.
  • SMS or voice codes: Confirming a number by code protects against third-party numbers, but it creates costs itself. Without limits, code delivery becomes a pumping target.
  • API and automations: Dial jobs from a CRM, Zapier flows or MCP tools run without anyone looking at the number. A stolen API key is enough for hundreds of calls.
  • Call transfers: Many voice agents hand over to a human on request. If the transfer number is changed to a premium-rate number, every caller is connected there at your expense.
  • Long call duration: Fraudsters keep the line open with recorded announcements. A polite AI agent without a time limit listens patiently – minute after minute.

The consequence: AI telephony needs the same protection as a traditional phone system, only applied more consistently, because no human sits in between anymore.

Common cost traps: premium numbers, international destinations, transfers

In Germany, these number ranges are particularly relevant for toll fraud. They have legitimate uses, but a business voice agent practically never needs to call them:

Number range (Germany)PurposeToll fraud risk
0900Premium-rate servicesHigh per-minute rates, paid out directly to the provider
0137 / 0138Mass-call services (e.g. televoting)Charged per call, easy to trigger at scale
0180 / 0181Service numbers and international virtual networksMore expensive than regular landline calls
019xOnline and special servicesRarely legitimate in B2B contact
0700Personal numbersCan forward to any destination, including expensive ones
118xxDirectory enquiry servicesHigh charges, often with onward connection
InternationalEspecially destinations with high termination ratesMain target of IRSF

The Bundesnetzagentur describes the pattern on its router hacking page: criminals gain access to a router or phone system and call international, (0)180 or (0)900 numbers. Often nobody notices until the provider flags unusual connections or the bill arrives.

Three traps are often overlooked:

  1. Mobile and premium ranges inside allowed countries: Even if a country is enabled, individual number ranges there can carry premium charges.
  2. Transfer and callback numbers: Often only the number the agent dials is checked – not the number it transfers to.
  3. Look-alike country codes: An international code that resembles a domestic area code slips through a visual check.

How do you prevent toll fraud? The checklist

The most effective toll fraud prevention is to enable only the destination countries you need, block special-rate numbers by default, check every number before dialing and cap costs hard with limits and prepaid credit. No single measure is enough on its own; only the combination makes billing fraud unattractive.

The checklist for businesses that use AI telephony or callback forms:

  • Destination country allowlist: Enable only countries you actually call. Twilio's documentation on geo permissions explicitly recommends enabling only the countries and low-risk number ranges you need.
  • Special-rate number block: Block 0900, 0137/0138, 0180/0181, 019x, 0700 and 118xx as well as international premium and shared-cost ranges. According to the Bundesnetzagentur, German providers block (0)900 ranges at network level free of charge on request.
  • Check transfers: Every number a call is transferred to goes through the same check as dialed numbers.
  • Rate limits: Caps per destination number, per IP address, per account and per day – for calls and for SMS codes.
  • Bot protection on forms: CAPTCHA solutions such as Cloudflare Turnstile, honeypot fields and checks for suspicious patterns.
  • Double opt-in by code: Call back only after the person has confirmed by SMS or voice code that the number is theirs.
  • Time windows and call duration: Calls only at fixed times and with a maximum duration.
  • Prepaid instead of an open invoice: Credit limits the maximum damage; once it is used up, no new calls start.
  • Account suspension on suspicion: Stop suspicious accounts automatically or manually before the damage grows.
  • Monitoring: Make unusual destinations, spikes and night-time activity visible; review invoices and call records regularly.
  • Harden access: Strong, unique passwords, API keys with restricted scopes, updates for routers and phone systems.

Risk β†’ measure: toll fraud at a glance

RiskTypical scenarioEffective measure
IRSF via international destinationsHacked account dials expensive foreign numbersCountry allowlist, block premium ranges
Premium-rate numbersAgent is told to call a 0900 or 118xx numberSpecial-rate number block in every country
Manipulated transferTransfer target is changed to a premium numberCheck transfer numbers like dial targets
Callback form abuseBots enter third-party or expensive numbersCAPTCHA, double opt-in by code, country restriction
SMS pumpingMass delivery of verification codesRate limits per number, IP and day
WangiriCallback to a missed lure numberNever auto-call back unknown international numbers
Stolen API keyAutomated dial jobs without oversightScopes, approval gate for call lists
Unlimited costsThe bill shows the damage weeks laterPrepaid credit, monitoring, account suspension

What should you do if you have been hit by toll fraud?

If you have been hit by toll fraud, immediately lock the affected accounts, rotate passwords and keys, dispute the bill and report the case to the authorities and the police. The faster you react, the smaller the damage.

For router and phone system hacking in Germany, the Bundesnetzagentur specifically recommends:

  1. Dispute the bill: Object to the charges with your provider without delay and request an itemized call record.
  2. Check direct debits: If the bill was paid by direct debit, you can have it reversed and pay only the undisputed amount.
  3. File a complaint: Report the case to the Bundesnetzagentur with the bill and call records. It can switch off numbers and ban billing and debt collection.
  4. Report to the police: Toll fraud usually involves a criminal offense.

For cloud telephony and AI platforms, also report the incident directly to the provider so it can lock the account, numbers and API keys. Then check that destination countries, transfers and limits are set correctly.

How anicall.io handles it

anicall.io is an AI phone assistant for businesses in the DACH region, operated by mysoftwarelab GmbH from Worms, Germany. Because our voice agents dial on their own, protection against billing fraud is built in and not optional:

  • Dial protection for special-rate numbers: Premium-rate, special, service and emergency numbers (e.g. 0900, 0137/0138, 0180/0181, 019x, 0700, 118xx) are never called – in any country. In addition, all numbers classified as premium or shared cost are blocked.
  • Destination countries per account: The default is Germany, Austria and Switzerland. Further countries can only be enabled from a fixed list (core EU, UK, USA, Canada).
  • Transfers are checked, too: The number for the live transfer to a human goes through the same check as every dialed number.
  • Demo callback only after code confirmation: The demo call on the homepage only happens once the phone number has been confirmed by SMS or voice code. Calls run only at fixed times and with caps.
  • Approval gate: Call lists from the API, CRM or MCP only start after a human approves them. There is also an account-wide do-not-call list.
  • Account suspension on suspicion: Accounts can be suspended if misuse is suspected.
  • Prepaid tokens instead of an open invoice: No base fee, no subscription. 1 token equals 1 cent net; inbound calls cost 13 tokens per minute, outbound calls to landlines 16 and to mobiles or abroad 62 tokens. When the balance is used up, no new calls start; a running call is limited to the remaining balance.

A calculation example from the pricing page: 300 inbound minutes per month plus one phone number come to 300 Γ— 13 + 180 = 4,080 tokens, i.e. €40.80 net. You know in advance what the worst case can be. For database, encryption and approvals, see Security & Privacy; how the outbound agent dials lawfully is explained on the AI setter page.

Frequently asked questions

What is the difference between toll fraud and IRSF?

Toll fraud is the umbrella term for billing fraud in telephony. IRSF (International Revenue Share Fraud) is its most common form: fraudsters drive calls to expensive international numbers and receive a share of the charges.

Who pays for toll fraud?

Usually the holder of the line or account the calls went through. Telephony providers such as Twilio explicitly make their customers responsible for all usage. That is why blocks and limits pay off before anything happens.

Is blocking 0900 numbers enough?

No. 0900 is only one of several expensive number ranges, and most IRSF runs through international destinations. You also need a destination country allowlist, a check of transfer numbers and limits.

How do I protect a callback form from abuse?

Combine bot protection (e.g. a CAPTCHA), confirmation of the number by SMS or voice code, country restrictions and rate limits per number, IP and day. Only call back once the number is confirmed, and limit time windows and call duration.

Is prepaid really a protection against toll fraud?

Prepaid does not prevent the attack, but it caps the maximum damage at the available credit. With an open invoice, toll fraud often only shows up weeks later – with prepaid, it stops when the balance runs out at the latest.


Test the protection yourself: Try the live demo on anicall.io: call the AI reception right in your browser or request a demo call – which only happens after you confirm your number by code. Then get started for free on anicall.io: the agent test in the browser costs nothing, and you only need tokens for real phone calls. The phone agents currently speak German.