What problem does it solve?
Smishing is phishing sent by text message. The messages remind people of an overdue bill, ask them to complete missing personal details or promise a refund, and they usually carry a link to a fake website that steals data or money. A text message arrives on the phone's own messaging app and often from a number the recipient has never seen, so the victim has little to judge it by.
Operators see what a single handset cannot: one number sending a remarkably high number of messages, or the same suspicious link appearing in many messages at once. Scammers adapt: Telstra says that if criminals guess a message with a suspicious link has been blocked, they change the characteristics of the message to avoid being blocked again. Proximus and Telenet group started blocking in autumn 2023 under StopPhishing, a federal project that in 2022 subsidized new anti fraud platforms at the two operators for text messages, email and voice calls.
- Proximus cites Febelfin figures under which fraudsters took nearly EUR 40 million from people in Belgium in 2022 through phishing of all kinds, including text messages, emails and social media posts.Proximus already stopped over 3 million suspicious text messages thanks to new platform (2023)
How does it work?
- Score every message in transit. The platform sits in the operator's messaging path and looks at patterns such as a number sending an unusually high number of messages, the same suspicious link spreading across messages, and the content of messages that look suspicious where the law allows it.
- Block above a threshold. Messages the model rates as phishing are stopped automatically and never reach the subscriber. A Telenet group fraud expert gives the example of one number sending up to 10,000 messages in a short time with a suspicious link.
- Retrain on new campaigns. Proximus says the algorithm can be trained further to recognize phishing messages better over time, and Telenet group describes its platform as self learning, so detection keeps up with new wording and sending patterns.
- Report large campaigns. Blocked volumes and sender patterns go to the operator's fraud team, and Telenet group files a complaint with the police when it spots a large scam campaign.
- Audience
- Back office
- Autonomy
- Autonomous
- Adoption
- Mainstream
- Channels
- SMS
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Interactions handled | Not pooled | 2.4 million to 98.4 billion | 4 | 4 organization |
Value drivers: Risk and loss reduction, Customer experience.
Indicative value
A mobile operator with 3 million subscribers
EUR 180,000 to EUR 3.6 million
Customer losses prevented per year
How this is calculated
Formula: subscribers * victimRate * averageLoss * preventedShare. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Subscribers protected subscribers, subscribers | 3,000,000 | 3,000,000 | The reference operator. |
| Share of subscribers who lose money to a scam text message in a year victimRate, fraction of subscribers | 0.001 | 0.003 | Editorial assumption, replace with your own fraud report data. |
| Average loss per victim averageLoss, EUR per victim | 300 | 800 | Editorial assumption, replace with your own fraud report data. |
| Share of those losses prevented by blocking preventedShare, fraction of losses | 0.2 | 0.5 | Editorial assumption, deliberately conservative. Proximus quotes a UK initiative that led to a 90% drop in smishing complaints, but a complaint count is not a loss count. |
What it leaves out: Customer losses only, and a rough order of magnitude. It leaves out the operator's savings on scam related contacts, the effect on customer trust and the cost of genuine messages that are blocked by mistake.
Who already uses it?
4 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Proximus
Belgium · Telecommunications · 2023
Proximus, the Belgian telecom operator, put a platform into use on 4 October 2023 that detects and automatically blocks smishing, phishing sent by text message, before the messages reach customers. The platform looks for patterns such as numbers that send a remarkably high number of messages and suspicious links, and Proximus says artificial intelligence lets the algorithm be trained further to recognize phishing messages better. The operator reported 3.2 million blocked messages in the first two weeks. The platform was developed with public funding under the federal StopPhishing project, and Belgian press reported in 2025 that Proximus and Telenet group filter text messages, emails and voice calls with detection systems trained with artificial intelligence.
- Interactions handled: 3.2 million, first two weeks after launch on 4 October 2023, suspicious text messages blocked
"Initial results are promising: in the first two weeks after implementation, as many as 3.2 million suspicious messages were blocked, which comes down to 229,000 per day on average."
Claimed by: organization
Telenet group
Belgium · Telecommunications · 2023
Telenet and BASE, the mobile brands of Telenet group, started using a self learning platform in autumn 2023 that detects and automatically blocks smishing, phishing by text message. The platform recognizes patterns such as numbers that send a very high number of messages and suspicious links being sent around. Telenet group paid half of the development and running cost, and the rest came from public funding. In VRT NWS reported in February 2024 that Telenet and Proximus had used new software with AI since October, and in 2025 a Telenet group fraud expert told VRT NWS that the operator's detection systems for text messages, emails and voice calls are trained with artificial intelligence.
- Interactions handled: 2.4 million, since the launch of the platform in autumn 2023, suspicious text messages blocked
"Since the implementation, 2.4 million suspicious messages were blocked, which comes down to 170,000 per day on average."
Claimed by: organization
Telstra
Australia · Telecommunications · 2022
As part of its Cleaner Pipes initiative, Telstra launched an SMS scam filter in April 2022 that scans text messages as they travel across its network and blocks malicious messages with suspicious links or telephone numbers before they reach the customer.
Telstra says the content is scanned by a machine that looks for suspicious patterns and characteristics, and that a small team in Australia reviews a sample of suspected messages, with the recipient's identity masked, to confirm a scam and adjust the technology to block future messages with the same criteria or near matches. Messages from banks and other businesses, government departments and Emergency Alerts are not scanned, and customers can opt out.
The filter product comes from an unnamed specialist telecommunications supplier and is operated by Telstra staff in Australia. Telstra reported more than 225 million scam messages blocked between April and December 2022.
- Interactions handled: 225 million, April to December 2022, scam text messages blocked
"Since then, we have continued to block millions of scam messages, with a staggering 225 million blocked since April this year."
Claimed by: organization
Bharti Airtel
India · Telecommunications · 2026
Airtel rolled out India's first network based, AI powered spam detection system for a telecom operator in September 2024, analyzing call and message patterns at the network level to flag spam and fraud in real time for mobile and Wi-Fi broadband subscribers, without an app to install. The service has since grown into a wider Safe Network program that also screens malicious links and warns customers being talked into sharing an OTP. Airtel reported a two year cumulative total of spam calls and messages identified by the system in September 2026.
- Interactions handled: at least 98.4 billion, cumulative since the system's September 2024 launch, reported September 2026
"According to Airtel, its AI-powered network security tools have identified more than 98.4 billion spam calls and 4.4 billion spam messages since the spam detection system was launched."
Claimed by: organization
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Message metadata from the SMS center, such as sender, volume and timing
- Message content and links, where the law permits inspection
- Confirmed scam reports from customers and from the national cybersecurity authority
- Lists of legitimate bulk senders such as banks and delivery firms
Systems to integrate
- Messaging core and SMS firewall for blocking in the path of the message
- Fraud team case tools for reviewing sender patterns
- Customer reporting channel for missed scams and wrongly blocked messages
- Exchange of sender intelligence with other operators and regulators
Complexity: Medium
The platform needs access to the messaging core (SMS center and messaging firewall) and a legal basis to inspect message content, which Belgian operators received through a change in the law. The Belgian platforms were built with public funding that covered half of the cost over three years, so the business case and the regulatory permission come before the model work.
- 1
Start with sender behavior
Block numbers that send an unusually high number of messages and links already known as malicious before you read message content.
- 2
Settle the legal basis
Agree with the data protection officer and the regulator what may be inspected, for how long data is kept and who may see it, then add content analysis for suspicious messages.
- 3
Register legitimate senders
Give banks, delivery firms and authorities a way to register sender names and links so their bulk messages are not blocked.
- 4
Measure blocked volume and harm
Track blocked messages per day next to scam reports per thousand customers, so a falling report rate confirms that blocking works.
Guardrails
- Block automatically only above a validated confidence threshold
- Messages to emergency services and from registered public authorities are never blocked
- Content inspection limited to messages that are already suspicious on behavior
- A dispute path for senders whose messages are wrongly blocked
KPIs to instrument
- Suspicious messages blocked per day, per category
- Scam reports per thousand customers
- Legitimate messages blocked by mistake and disputes upheld
- Time from first appearance of a campaign to its block
Human in the loop
Fraud analysts set thresholds, review new campaign patterns and handle disputes from senders. Customers can report missed scams, and those reports feed the next round of training.
Common failure modes
- Legitimate bulk senders blocked
- Banks and delivery firms send many messages with links. Without a sender register, genuine one time codes and parcel notices get stopped.
- Blocked volume reported as the result
- Millions of blocked messages say little about harm prevented. Pair the count with scam reports and reported losses.
- Scammers move channel
- Scammers shift to email, voice calls and messaging apps once text messages get harder. Coordinate protection across channels.
What are the risks and rules?
EU AI Act
Minimal risk
Scoring and blocking messages is not listed in Annex III and is not a prohibited practice under Article 5. The system does not interact with people or generate content, so Article 50 does not apply. A message filter protects subscribers from fraud and is not a safety component in the management and operation of critical digital infrastructure under Annex III point 2 (Recital 55 excludes components used solely for cybersecurity). It is governed mainly by telecom and data protection law.
Rules that apply
Guidance
- Scam calls and messages (Ofcom, Europe). UK regulator hub on scam calls and messages, including its statement on tackling scam calls from abroad, which covers its calling line identification guidance on how providers should process calls from abroad that present a UK mobile number.
Controls to put in place
- Documented blocking policy aligned with national telecom law
- Privacy notice and legal basis for inspecting message content
- Monitoring of false positives and sender disputes
- Regular review of thresholds against new scam campaigns
Frequently asked questions
- How many scam text messages do operators block?
Proximus reported 3.2 million suspicious messages blocked in the first two weeks after its platform went live on 4 October 2023, and Telenet and BASE reported 2.4 million after launch in autumn 2023. VRT NWS reported in September 2025 that Proximus and Telenet group together blocked 37.3 million text messages in 2024, and 10.75 million text messages, 165 million emails and 16 million incoming voice calls between 1 January and 6 July 2025. Telstra reported more than 225 million scam text messages blocked between April and December 2022.
- Is this the same as scam call blocking?
No. Call blocking labels and stops scam calls on the voice network, while this use case works on text messages in the messaging network. Operators often run both, and VRT NWS reports that the Belgian systems also filter voice calls and email.
- Why do operators need a change in the law for this?
Reading message content is restricted by the confidentiality of communications under the ePrivacy Directive. A Telenet group fraud expert told VRT NWS that a change in the law lets its systems look at the content of messages that already look suspicious.
How to cite this page
Blits.ai AI Use Case Library, "AI blocking of phishing and scam text messages in the operator network", last verified 11 November 2026, https://www.blits.ai/ai-use-cases/sms-phishing-and-scam-message-blocking. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 11 November 2026: First published