A Cape Town-based technology startup, Untapped AI, has filed two patent applications in the United States for technology designed to verify the origin of phone calls and protect users against scams involving AI-generated voice clones.
The company said the technology, which was tested but is not yet commercially deployed, is intended to prevent criminals from impersonating bank officials, employers and family members to defraud unsuspecting victims.
Untapped AI, which develops artificial intelligence voice agents for businesses, has offices in Cape Town and Johannesburg. Its investors include Yaxxa, a business telecommunications and contact-centre provider, whose chief executive, Heinz Vollmer, confirmed the investment.
Untapped AI director Lloyd Matthew said the technology has two components, both involving a cryptographic key that changes with every call. The key is either exchanged at the network switch when a call connects or transmitted within the call audio as a signal that cannot be heard by the participants.
The second component examines the sound’s frequency, gaps and other signal markers to verify the integrity of the conversation rather than simply comparing a caller’s voice with a recorded sample.
“We verify where the voice is coming from, which means the source of the call,” Matthew said.
He compared the system to a bank card security mechanism, explaining that a criminal using a deepfake voice would still need to pass the technology’s authentication process, just as someone using a stolen bank card would need the correct PIN.
However, the company has not made the technology available to the public. Matthew said protection would only begin once mobile network operators or businesses deployed the system at both ends of a call.
Untapped AI has started discussions with a major mobile network operator but declined to identify the company. Matthew said deploying the technology through a network could extend protection to all its subscribers.
The development comes amid growing concerns about digital fraud. South Africans lost R2.4-billion to digital banking crime in 2025 across more than 110,000 reported incidents, according to the South African Banking Risk Information Centre (Sabric).
Sabric, however, said it could not determine how much of those losses originated from phone calls. It described AI voice cloning as an emerging threat but said available evidence did not indicate that it had surpassed conventional social engineering scams in South Africa.
The risks have already become evident internationally. In February, fraudsters reportedly stole €95-million from Italian private bank Fideuram, part of Intesa Sanpaolo, after using a fake WhatsApp message and a phone call featuring an AI-cloned voice of a senior lawyer. More than half of the money has since been recovered.
Despite the potential benefits, experts have raised questions about the technology’s transparency and effectiveness.
The company’s press release describes the system as authenticating the voice on a phone call, while Matthew’s written responses to TechCentral said the patents do not analyse the voice itself. He said both descriptions were accurate but acknowledged that the wording could have been clearer.
Matthew also declined to disclose the technical details or publish test results, saying the company wanted to avoid revealing its security process publicly.
See also: Meta Suspends Trump Protest Movement’s Instagram Account, Then Apologises
Professor Martin Olivier of the University of Pretoria’s computer science department said the information available was insufficient for him to fully assess the claims. He noted that cryptographic systems widely trusted today, including AES and RSA, rely on publicly available technical specifications.
Olivier also warned that audio compression used by telecommunications networks could distort signals, potentially affecting how such a system performs.
The proposed technology differs from the Stir/Shaken call authentication framework used by US networks, which verifies the authenticity of the caller’s number rather than the audio itself. Matthew said Untapped AI’s approach does not rely on the number displayed on a recipient’s phone.
Andries Maritz, principal product manager at financial technology company Entersekt, said embedding a signal within call audio could be a significant development, provided the system works as intended.
However, he warned that fraudsters could bypass network-based authentication by using internet-based calling services or other routes. Banks would therefore still need additional fraud prevention measures.
Sabric also cautioned against treating call authentication as a complete solution, noting that criminals could shift to WhatsApp, text messages, email and social media as security measures improve.
Banking technology executive George Chinomona, speaking in his personal capacity, said AI was making existing fraud techniques more effective and warned that banks could no longer rely on voice alone to establish a person’s identity.
Matthew said Untapped AI hoped governments would introduce policies shifting responsibility for losses from fraudulent calls away from customers and towards businesses. He added that the company would be willing to share its code with the wider market, including competitors, if such a policy were adopted.
Until the technology is deployed, security experts advise people who receive calls requesting money, PINs or one-time passwords to hang up and call back using a number they already know or have saved.
About the Author
Cecilia Attah is a tech analyst with a degree from Benue State University. She covers tech news and startups at TechRegard with a focus on how technology is transforming Africa and shaping the global landscape.
