Rostelecom has disclosed to suppliers the requirements for a large-scale AI platform for analyzing voice calls for millions of subscribers. CNews writes about this.

This refers to a platform for analyzing incoming and outgoing calls for fraud in real time, with integration with the GIS Antifraud. Applications from developers are accepted until October 15, 2026. Based on the results of the request, the company will not be obliged to conclude a contract.

What exactly is required

The platform must process voice traffic, recognize speech, determine emotions, identify behavioral anomalies, identify the speaker, detect deepfakes and speech synthesis. The quality of fraudulent call recognition is stated at 99%. Model retraining is to be done at least once every two weeks.

Scale — traffic throughput for 3 million subscribers, at peak up to 30 thousand simultaneous calls and 1200 new calls per second. Placement sites — Moscow, St. Petersburg, Nizhny Novgorod, Novosibirsk, Yekaterinburg and Rostov-on-Don.

The main intrigue — messengers

In the list of technologies where the platform should operate, Rostelecom mentioned not only the usual 2G/3G, VoLTE and VoWiFi, but also the Max and VK messengers. At the request of CNews, VK studied the documentation and stated that the requirements put forward are “technically unfeasible in our products.”

Experts explain: calls in messengers are encrypted, and the operator does not see them. Such traffic can only be analyzed through integration with the platforms themselves, but how exactly this is arranged is not disclosed in the technical specifications. At the same time, the law obliges only telecom operators and banks to compensate for damages; this obligation does not extend to messengers.

What Rostelecom says

The company stated that the request is for informational purposes and does not imply any procurement obligations. They noted that they regularly study the market for new solutions to combat fraud, and a decision on the feasibility of procurement will be made after analyzing the information received.