A call comes in from a number you don’t recognise. Ten years ago, the worst outcome was wasted time. Today, it might be a fraudster who knows your bank, your recent transactions, and your name. That knowledge is precisely what makes them convincing.
Scam calls are not a new problem for the telco industry, but AI has dramatically increased their sophistication. Previously, scams were clearly automated, and it was easy to recognise the robotic tone or the obvious script. Today, fraudsters generate personalised approaches at scale, adapt their tactics in real time, and work from data that makes them sound legitimate. What was once a nuisance has become a convincing threat.
The numbers reflect this shift. One in four adults is harassed by scam calls daily, and 73% of UK adults have reported being a target of scams, turning a trusted channel into one of anxiety and suspicion. In 2025, global telecom losses from fraud exceeded $80 billion. Many have tried to solve this with applications or device-level integrations. But no one is better positioned to fight back than the telcos themselves.
Why the Network Is the Right Place to Act
When it comes to fraud prevention, not all solutions are equal. Device-based technologies such as apps, settings, and consumer reporting tools rely on limited crowd-sourced data. It is useful, but it offers narrow protection.
The network is different. Analysing real-time network data with AI, operators can act before the call reaches your phone. A call that looks legitimate to the user can carry dozens of warning signs: call frequency, roaming status, routing anomalies. The network sees these indicators and can alert you instantly when something is wrong. Acting on those signals, at speed and at scale, is exactly what AI is built to do.
There is a real trade-off here. Device-level tools sit where the consumer already is: pre-installed, familiar, requiring no decision from anyone. That distribution advantage is significant, and it is the reason device-based screening has become the default for most users. But distribution is not detection. The network has the signals; the challenge for operators is meeting consumers where they already are, rather than asking them to go somewhere new.
This is where telecoms’ natural advantage becomes a powerful tool. By extending capabilities that have always existed, such as call authentication, network monitoring, and traffic analysis, AI can process more signals far faster than any other system, giving operators fraud detection that works without requiring consumers to install anything or change how they use their phone.
The result is network-level protection that happens before the phone rings rather than after the damage is done.
Why Collaboration Makes the AI Smarter
Building AI that works at the network level sounds straightforward in principle. In practice, the accuracy of the model depends almost entirely on the quality and diversity of the data it learns from.
Fraud does not respect borders. A campaign targeting consumers in one country may use techniques first observed elsewhere, evolving constantly based on what has worked in other markets. A model trained on one operator’s network will learn the patterns it has seen and struggle with those it has not. For network-level AI to be effective against adaptive threats, it needs training data that reflects that same breadth.
This is not a new idea. Weather forecasting became more accurate when nations agreed to share atmospheric data across borders. The models improved not because the algorithms changed, but because the training data became representative of global conditions.
But the meteorological comparison also shows what makes this hard. Those agreements took decades, and weather data is not commercially sensitive. Network data is among the most tightly regulated data there is, and moving it across borders runs into GDPR, national security rules, and a patchwork of local requirements that were not designed with collaborative AI in mind.
The hard part, in other words, is not the model. It is the governance: what gets shared, in what form, under whose authority, and with what protections. Federated approaches, where models learn from data that never leaves its home jurisdiction, offer one route. The point is that the technical case for collaboration is settled, and the industry’s task is to build the frameworks that make it possible.
Telecom-native AI that learns from billions of call patterns across multiple networks, geographies, and regulatory environments will ultimately perform more accurately. That requires operators to work together.
Restoring Trust, One Call at a Time
The rise of scam calls has cost more than money. It has made people afraid to pick up the phone. Unknown numbers go unanswered. Important calls are missed, business outreach goes unheard, and the default assumption about an incoming call has shifted from curiosity to suspicion.
Rebuilding that confidence is a technological challenge, but one that the telecom industry is well-placed to meet. The network has always carried the signals. AI is what finally makes them useful, not to sell consumers something new, but to give them back something they used to have: the ability to answer the phone without thinking twice.
