MTS has carried out a large-scale update of its anti-fraud platform "Zashchitnik," integrating three specialized AI models into it. The result is impressive: the system's average response time to a suspicious call has dropped from 30 to 15 seconds, and the accuracy of detecting complex multi-stage fraud schemes has tripled. This is not just a cosmetic improvement, but a qualitative leap in the fight against phone fraud.
How the updated system works
The key change is the simultaneous analysis of a call by three neural networks. Every second, they process more than 1,100 parameters, including not only the technical characteristics of the connection, but also the context of the conversation, the speed of the interlocutor's speech, and other behavioral markers. This comprehensive approach makes it possible to identify not only obvious spammers, but also fraudsters who constantly change their pressure tactics and move on to multi-stage attacks.
Special emphasis is placed on multi-step schemes, when attackers make a series of calls to a single subscriber. To improve accuracy, the models were retrained on new data and supplemented with more complex language algorithms. This is critically important, since phone calls are often just the first link in a chain leading to money transfers, including in cryptocurrency.
Scale of the threat and other schemes
From January to July 2026, the service detected and blocked more than 1.5 billion spam and fraudulent calls. The figure is colossal and clearly demonstrates how massive attacks on subscribers have become. Reducing response time is critical: the faster the system interrupts contact, the less time the victim spends under psychological pressure.
Other attack vectors are being recorded in parallel. Chatbots with "poisoned data" lure buyers to fake stores, fraudsters offer to buy top coins at a fixed Central Bank rate through phishing gateways, and drainers disguise themselves as partner investment programs and empty wallets via QR codes. On Telegram, infostealers that steal passwords and session tokens are distributed under the guise of secretary bots.
Such scams often end in criminal cases. In Moscow, police detained a courier who collected cash from pensioners, converted it into digital assets, and sent it to accomplices — the damage from two incidents exceeded 6 million rubles.
My take: Accelerating the response of anti-fraud systems is a real-time arms race. Fraudsters are increasingly using AI to personalize attacks, and operators have to respond symmetrically. The fact that three models work in parallel is the right approach: it reduces the likelihood of false positives and increases adaptability to new tactics. However, subscribers should not let their guard down: technical protection is just one of the barriers, and basic digital hygiene remains critically important.