Beeline's AI voice personas for scam calls

Russian mobile carrier Beeline said on 6 October that its anti-fraud platform has routed 19.1 million scam calls to a set of AI voice personas since the start of 2026. Beeline calls them its "cyber family", and according to the press release, scammers have spent about 140,000 hours talking to them, close to 6,000 days. The service runs only on Beeline's network in Russia, and all figures are self-reported without independent verification.

Beeline began small-scale tests of a "cyber granny" in November 2025 with a single voice, and the release at the time said suspicious calls to older female customers would be routed to it. The system analyses what the scammer says, then a large language model generates a reply from the conversation's context and turns it into speech. A teenage persona was added in June 2026, and in August Beeline announced ten voices in total, with almost all scam calls on its network now sent to them. The AI picks a voice based on the number being called, and a scammer who calls the same number again within five hours reaches the same persona.

Beeline's June 2026 release says the service is part of the free basic tier of its "My Security" (Моя безопасность) service. The chief executive said on 6 October that these calls are also a large body of material for retraining the models. The releases do not say whether the protected customer's phone rings, whether customers can switch the service off, or what happens when a real person's call is misrouted to the AI.

Perspective

According to Beeline's releases, its anti-fraud platform decides which calls look suspicious and where to send them, without publishing the criteria. The point of stalling scammers is to use up their time so they make fewer calls to real people. O2 in the UK launched a similar AI granny, Daisy, in November 2024, and its press release says Daisy kept scammers on the line for up to 40 minutes at a time, without saying how many calls it took.

Both the decision and the routing happen on the carrier's side, and as of 12 October Beeline has not published how misrouted calls are handled. The releases say the calls are used for retraining but not whether recordings or transcripts are kept, for how long, or on what legal basis. South Korea offers a contrast. According to the government's policy news site, the Ministry of Science and ICT said in July 2025 that a scammer's voice counts as sensitive information under Korea's Personal Information Protection Act, so the carrier KT needed a regulatory sandbox exemption (a trial scheme that temporarily exempts a limited use from existing rules) to use such voice data without consent for an AI service that detects scams during calls.

In the three places we checked, South Korea, Singapore and Taiwan, we found no carrier offering AI personas that answer scam calls as of 12 October, only call warnings, filtering and on-device detection. In February 2026 Korea's science ministry urged people to use AI scam-call detection from Samsung and the three mobile carriers, noting that call analysis runs on the phone itself rather than on outside servers. Singapore's ScamShield combines an app that helps people check, block and filter scams with a 24/7 helpline at 1799. In Taiwan, Chunghwa Telecom introduced a spoken warning in July 2023 for international calls showing a "+886 9" prefix on landlines.

Wherever you are, the advice from Taiwan's Criminal Investigation Bureau, relayed by the Ministry of the Interior, is to hang up first and then verify through an official channel. In Taiwan, a caller who asks you to dial 165, the anti-fraud hotline, without hanging up first is a scammer. The 165 hotline and the 165 website take enquiries and reports. Carrier-side call interception like Beeline's only handles incoming phone calls, so in our view scams that start in messaging apps still need checking yourself.