The call sounds like your mum. She is upset, she needs money now, and she cannot explain why. Voice-cloning tools have made that kind of call cheap to fake: in tests published in May 2023, McAfee researchers found that three to four seconds of recorded speech was enough for one free tool to produce a clone that they estimated was an 85% match for the real voice.
In the UK, these calls are already common enough to register in surveys. Our 43-second explainer above shows how the scam works and the simple defence that beats it. This article sets out the figures behind it, the limits of that data, and what you can do this week.
In brief
- What happened: 28% of UK adults told a Starling Bank survey they think a voice-clone scammer targeted them in the past year, and 46% had never heard of the scam.
- Why it matters: a few seconds of public audio can be enough to imitate a relative, and a family safe phrase agreed in person is a defence a clone cannot copy.
- What’s unknown: how many UK victims lose money to cloned voices specifically, since none of the sources used here separates them from other impostor scams.
Over a quarter of UK adults think they have been targeted
In August 2024, Starling Bank and Mortar Research asked 3,010 UK adults about voice-cloning scams. The bank published the results on 18 September 2024: 28% said they think they had been targeted by an AI voice-cloning scam in the past year, while 46% had never heard of such scams. Only 30% said they would confidently know what to look out for.
These are self-reported answers in a survey commissioned by a bank, not confirmed cases, and some people may be describing calls that were never cloned. But the awareness gap is the more striking number. Nearly half of the adults asked did not know this kind of scam existed.
Starling describes the pattern in plain terms. A short clip of someone’s voice, which “can easily be captured from a video someone has uploaded online or to social media”, is used to copy it. The scammer then works out who the person’s family members are and stages “a phone call, voice message or voicemail” asking for money that is needed urgently. Starling also warns that scammers can spoof a number, so a call can look as if it comes from someone you know.
The bank ran the campaign in support of the government’s Stop! Think Fraud campaign. Lord Hanson, the Home Office minister responsible for fraud at the time, called it “practical advice about how to stay protected from this appalling crime”.
US figures show what impostor scams cost
The Starling survey does not measure money lost. For losses, the clearest large-scale data on impostor scams comes from the US. In June 2026 the Federal Trade Commission said that people reported losing $3.5bn to impostor scams in 2025, nearly three times the reported figure for 2020, and that almost one in three fraud reports it received were about impostors.
That total covers every kind of impostor, from fake bank staff and government officials to people pretending to be a friend or relative. It is not a voice-clone figure. It is also only what people chose to report. In testimony to Congress in March 2026, the FTC said reported fraud losses “are just a fraction” of actual losses, because not every victim reports.
A shared secret beats a perfect copy
The defence works because of what a clone cannot copy. Software can imitate how someone sounds from a short clip. It cannot know a phrase that your family agreed face to face and never wrote down online. Starling’s Safe Phrases guidance describes it as “a previously agreed phrase that you and your inner circle can use to verify if you’re truly speaking to them”. It should be simple but random, easy to remember, different from your passwords, shared in person and never saved digitally, and changed every now and again, or straight away if you think it is no longer private.
McAfee’s 2023 report gives the same advice in different words: set a “codeword” with children, family members or close friends, and ask for it whenever they call asking for help. If the caller cannot give it, both sources point to the same next step. Starling says to end the call, make no payment and contact the person another way. McAfee says to “hang up and call the person directly”. In practice that means ringing back on the number you already have for them, not the one that called you.
The approach has a useful property: it does not depend on spotting the fake. As cloned voices improve, listening for odd pauses or a flat tone becomes a weaker test. A safe phrase turns the question from “does this sound like Mum?” to “does this person know what Mum knows?”, which is much harder to fake. In the UK, anyone who thinks they have been caught can call 159 to reach their bank, a number Starling highlights on its campaign page.
Banks and platforms are building the second line of defence
A family phrase is the first line, but the money usually moves through a bank. Voice-clone requests are a form of authorised push payment fraud, where victims are tricked into sending the money themselves. The Financial Conduct Authority is consulting on new guidance for banks and payment firms on APP fraud. Its draft lists warning signs such as customers “using increasingly desperate tones” and proposes updating case studies to cover deepfakes. That consultation closes on 24 December 2026.
Technology companies are adding filters at the point of contact. Microsoft plans AI phishing detection in Teams that reads a conversation’s intent, and Google has added scam safeguards to Messages and Android. From a compliance point of view, those controls and a family’s own check are complementary. A bank can question an urgent payment to a new payee, but only the family knows the phrase.
The same technology can give a voice back
Voice cloning is not only a tool for fraud. In October, Baroness Campbell of Surbiton became the first peer to address the House of Lords using an AI recreation of her own voice, built with ElevenLabs software from old recordings. The same ability to learn a voice from audio is what makes the scam possible, which is why the defence relies on a secret, not on the sound.
The fix takes minutes. Agree a phrase with the people who matter, in person, and keep it off your phone and social media. The next UK milestone is the FCA’s final text of its fraud guidance, which follows the consultation’s close on 24 December.
Sources
- Starling Bank, “AI voice cloning scams could catch millions out”, 18 September 2024
- Starling Bank, “Safe Phrases: Stay safe against AI voice cloning”
- McAfee, “Beware the Artificial Impostor”, May 2023 (archived copy)
- US Federal Trade Commission, “FTC Data Show People Reported Losing $3.5 Billion to Imposter Scams in 2025”, 15 June 2026 (archived copy)
- US Federal Trade Commission, prepared statement to the Joint Economic Committee, 25 March 2026
Related stories
- FCA Proposes Financial Crime Guide Update on APP Fraud and AI Tools
- Microsoft Plans AI Phishing Detection in Teams From November
- Google Messages Gets an “Enhanced Protection” Mode as Android Adds New Scam Safeguards on a Staggered Rollout
Follow @tsnmediahub on X for breaking AI news as it happens.
Follow on X
