MTS has introduced a new AI model to search for fraudulent calls in its Defender service. The time it takes to identify and block unwanted calls has been significantly reduced: according to the developers, an attacker can "live" in the carrier's network for no more than 30 seconds on average.
"Defender" is not a separate application, but an internal MTS service that protects users from fraudsters. It now learns to more effectively determine the intentions of the person on the other end of the call.
After updating the AI model, the average time it takes to identify and block unwanted calls has been reduced from 3 minutes to 30 seconds, with the total number of conversations processed by the AI reaching 200 million per day.
Defender analyzes phone numbers for over 200 indicators of possible fraudulent activity. For example, it measures the number of calls placed, the ratio of incoming to outgoing calls, the percentage of successful connections, and other criteria.
Three AI models are used in conversations between users and potential fraudsters, evaluating over 1,000 call parameters per second and identifying anomalies in mobile traffic. According to MTS, the updated service allows fraudsters to be identified with 95.17% accuracy.
In addition, the anti-spam model has been updated. MTS big data experts have trained the relevant components of Defender. Thanks to the decomposition and optimization of machine learning algorithms and the use of large language models (LLMs) to analyze voice content, the service has begun to identify and block advertising calls 30% faster.
From January to June 2025, the Defender service blocked 1.5 billion fraudulent calls, a 6% increase compared to the same period in 2024.
