Phone carriers are teaching AI to sound exactly like you
A new kind of phone carrier AI assistant is moving beyond robovoices and into something more unsettling: AI voice clone calls that sound like you. REALLY, a T-Mobile MVNO, is developing “Clone,” an AI assistant that learns your voice, speaking style, and communication preferences so it can answer calls on your behalf. Once trained, the system can pick up when you are busy or simply don’t want to talk, handle the conversation in real time, then send you a summary of what happened. The company pitches it as a way to offload tedious tasks—such as rescheduling appointments, dealing with customer support, or filtering unknown numbers—so users can save time and avoid call anxiety. Because the assistant is built directly into the phone carrier, it blurs the line between helpful automation and deep integration into some of the most sensitive data you own: your voice and your phone number.
The convenience pitch: outsourcing your most annoying calls
Proponents of phone carrier AI assistants emphasize convenience. Clone, the AI assistant from REALLY, is designed to jump into low-priority calls so users can focus on conversations that actually matter. Instead of sitting on hold with customer care or repeatedly calling hotels to confirm bookings, subscribers can let an AI voice impersonation system do the tedious work. The assistant can also filter spam and scam calls, even intentionally wasting scammers’ time and reporting how long it kept them occupied. For people who dread phone calls or feel drained by everyday admin, this level of automation feels like a natural extension of tools that already summarize emails, manage calendars, and draft messages. Plans for the MVNO’s service start at USD 50 (approx. RM230) per month, with voice-clone features currently in beta, signaling that carriers see AI-driven call handling as a premium, built-in feature rather than a niche add-on.
Voice cloning privacy risks: from creepy to dangerous
Behind the convenience lies a serious set of voice cloning privacy risks. To make AI voice clone calls sound authentic, carriers need detailed recordings of your voice and patterns of how you communicate. That data can be stored, processed, and potentially repurposed. Critics warn that once your voiceprint sits in a carrier’s systems, it becomes a high-value target for hackers and a tempting asset for advertisers and third parties. Past concerns about carriers—such as allegations that T-Mobile recorded user screens—fuel skepticism about granting them even deeper access. AI systems themselves are not fully battle-tested for long-term security; researchers have already shown how AI tools can be exploited to access private data like calendars. If a voice clone or its training data is compromised, attackers could generate convincing deepfake phone calls that bypass voice-based security checks, impersonate you to services, or manipulate friends and family.
Deepfake phone calls and the new fraud threat landscape
The same technology that lets a phone carrier AI assistant cancel your subscription could power highly convincing deepfake phone calls. If strong safeguards are missing, scammers could combine stolen voice samples with carrier-grade synthesis tools to impersonate victims with alarming accuracy. Unlike text-based phishing, voice impersonation security failures exploit emotional cues—tone, urgency, and familiarity—to trick targets into sharing passwords, one-time codes, or financial details. Family emergency scams and fake customer support calls become harder to spot when the caller sounds exactly like someone you trust. Because these AI systems are integrated with real phone numbers and networks, fraudulent calls can appear both technically and socially legitimate. Security experts worry that carriers are racing to deploy voice clones without standardized protections, such as strict access controls, robust audit trails, and clear technical limits on how cloned voices can be used or exported.
What users should demand: opt-in, transparency, and limits
As AI voice clone calls move from experimental to mainstream, users need more than marketing promises. Clear, meaningful opt-in is essential: no one’s voice should be cloned by default, and enrollment should spell out exactly what data is collected, how long it is stored, and who can access it. Carriers should disclose whether voice data is used to train broader models, shared with partners, or monetized through advertising. Users also need intuitive controls to pause, delete, or fully revoke their voice clone, along with logs showing when the AI answered calls and what it did. Strong authentication—beyond simple voice recognition—should protect any features that act “on your behalf,” especially where payments, account changes, or identity verification are involved. Ultimately, the promise of a phone carrier AI assistant is only worth embracing if its design respects privacy, constrains misuse, and keeps users firmly in control.