AI Voice Scam Calls: How Fake Family Emergencies Work

What is an AI voice-cloning scam?

An AI voice-cloning scam is a fraud in which a criminal uses software to imitate the voice of someone you know and then phones you with a made-up emergency. The most common version is the family emergency call: a “relative” is in trouble and needs money immediately.

The US Federal Trade Commission (FTC) warned on 20 March 2023 that a scammer can clone a relative's voice using a short audio clip taken from content posted online plus a voice-cloning program ^[1]. The FBI's alert I-120324-PSA, dated 3 December 2024, says criminals generate short audio clips of a loved one's voice to pose as a relative in a crisis and ask for urgent money or a ransom ^[2]. In its 2025 annual report, the FBI's Internet Crime Complaint Center (IC3) names grandparent or “distress” scams as the type in which voice cloning mimics a loved one in distress ^[3].

The European Parliament's research service (EPRS) assessed in October 2025 that AI voice cloning is the most convincing and hardest-to-detect way generative AI is used to create deepfakes for fraud ^[6].

What do the key terms mean?

• AI voice cloning is software that generates speech imitating a specific person's voice from sample audio.
• Deepfake is audio, images or video created or altered by AI so that it looks or sounds like a real person.
• Vishing (voice phishing) is fraud carried out by phone call or voice message.
• Family emergency scam, also called a grandparent or “distress” scam, is a fraud in which the caller poses as a relative in trouble and asks for urgent money.

How does a fake family emergency call work?

The pattern described by the FTC and FBI has four steps.

1. Collect audio. The criminal takes a short clip of the person's voice from content posted online ^[1].
2. Clone the voice. A voice-cloning program turns that clip into a voice that can say new sentences ^[1].
3. Call with a crisis. The criminal poses as a relative in trouble and asks for urgent money or a ransom ^[2].
4. Collect payment. The FTC says requests to pay by wire transfer, cryptocurrency or gift cards are signs of a scam ^[1].

The pressure is the point. A frightened listener is less likely to stop and check.

Why doesn't caller ID prove anything?

The EPRS says scammers use VoIP, internet-based calling, so that calls appear to come from legitimate numbers ^[6]. A familiar or official-looking number on the screen is therefore not evidence of who is calling.

Why do some scammers move the conversation to a messaging app?

In the FBI's December 2025 update on the impersonation of senior US officials, the actors in most cases contact victims by text message and ask that the conversation be moved to encrypted messaging apps such as Signal, Telegram or WhatsApp ^[7]. That case involves officials rather than family emergencies, but the tactic is relevant. The EPRS notes that monitoring is more complex on encrypted services such as WhatsApp, as no monitoring is permitted there, which limits what networks can detect ^[6].

Is it only grandparents who get these calls?

No. IC3 says the scam is evolving to mimic other family members or close friends in different emergencies ^[3]. The FBI also says criminals have used AI-generated audio to gain access to bank accounts ^[2], so the risk is not limited to emergency calls.

How much audio do scammers need to clone a voice?

The FTC says only a short audio clip is needed ^[1]. Starling Bank, citing its September 2024 campaign, said a clone can be made from as little as three seconds of audio ^[8][9]. That is a claim from a bank's campaign, and none of the sources used here independently tests it. Treat it as a warning about how little is needed, not as a measured threshold.

Is there enough public audio to copy?

The EPRS cites a finding that 53% of people share their voices online ^[6]. Videos, voice notes and public social media posts all count as potential source material. The FBI therefore advises limiting online content of your image or voice and making social media accounts private ^[2].

Are the cloning tools restricted?

Not consistently. In March 2025 Consumer Reports examined voice-cloning products from six companies: Descript, ElevenLabs, Lovo, PlayHT, Resemble AI and Speechify. It found that only Descript and Resemble AI took steps to combat misuse, while the others required only that users tick a box confirming they had the right to clone the voice ^[5]. The study shows how low the barrier can be on some commercial tools. It does not show that scammers used those particular products.

How common are AI voice scams, and how much money is lost?

Nobody has a clean count of voice-cloning scams alone. The best official figures cover wider categories, so read them as context rather than as a tally of cloned-voice calls.

What does the FBI's 2025 data show?

IC3 received 1,008,597 complaints in 2025, with reported losses of $20.877 billion, a 26% increase in losses from 2024 ^[3]. Of those, more than 22,000 complaints (22,364) reported AI-related information, with adjusted losses above $893 million ^[3]. For grandparent or “distress” scams specifically, victims claimed losses of over $5 million ^[3].

The AI-related total is dominated by other fraud types. IC3 reported over $632 million in investment complaints with an AI nexus (against more than $8 billion in total investment losses), over $30 million in business email compromise involving AI, over $19 million in confidence and romance scams with a likely AI nexus, and almost $13 million in AI-involved employment scams ^[3]. IC3 adds that many victims may not realise how much AI is involved ^[3].

What does the FTC data show?

The FTC reported on 15 June 2026 that people reported losing $3.5 billion to imposter scams in 2025, nearly three times the 2020 level ^[4]. Nearly one in three fraud reports were about imposter scams, the most reported fraud category ^[4]. Total reported fraud losses were about $16 billion, the highest on record and about 25% more than in 2024 ^[4]. The FTC's 2024 Consumer Sentinel Network Data Book counted 845,806 imposter scam reports for 2024 ^[10]. Imposter scams include many frauds that involve no AI at all.

How badly are older adults affected?

Complainants aged 60 and over filed 201,266 complaints with IC3 in 2025, up 37% from 2024. They reported $7.748 billion in losses, up 59%, with an average loss of $38,500, and 12,444 of them lost more than $100,000 ^[3]. In an August 2025 press release summarising its data spotlight, the FTC reported that combined losses reported by adults 60 and over who lost more than $100,000 to impersonation scams rose eight-fold, from $55 million in 2020 to $445 million in 2024 ^[11].

Why do surveys and complaints disagree?

Starling Bank's survey of more than 3,000 UK adults, run with Mortar Research in August 2024, found that 28% thought they had been targeted by an AI voice-cloning scam in the past 12 months, and only 30% said they would confidently know what to look out for ^[8]. CNN reported the survey as finding that more than a quarter of respondents had been targeted ^[9]. The survey is self-reported and run by a commercial firm. IC3's figure is based on complaints, which depend on victims reporting and recognising that AI was involved. The two measure different things, and the gap probably reflects under-reporting, unclear AI attribution and survey wording. No source used here measures the true rate.

Can you tell a cloned voice from the real person?

Often not. The official advice is mixed, and that itself tells you something.

• The FBI advises listening closely to tone and word choice ^[2].
• The same agency warned in May 2025 that AI voice cloning can sound nearly identical to a known contact ^[12].
• The EPRS says detection by machine learning and forensic analysis has numerous limitations for voice calls ^[6].
• In the Starling Bank survey, only 30% of respondents said they would confidently know what to look out for if targeted ^[8].

The practical conclusion is to rely on verification rather than on your ear. Listening can be a weak backup, but a convincing voice proves nothing.

What clues are still worth noticing?

The clues are in the situation, not the sound. Watch for an urgent request for money or a ransom ^[2], a demand to pay by wire transfer, cryptocurrency or gift cards ^[1], and a request to move to an encrypted messaging app (seen in the senior-officials campaign) ^[7]. Any of these should trigger the same response: stop and verify through another channel.

How do you protect yourself and your family from voice cloning?

The sourced advice from the FTC and FBI comes down to verifying through a channel the caller does not control.

1. Hang up and call back on a number you know. The FTC says not to trust the voice and to call the person using a number you know is theirs ^[1]. For calls that claim to come from a bank or organisation, the FBI advises hanging up, researching the real contact and calling that number directly ^[2].
2. Use a second route. If you cannot reach the person, the FTC advises contacting them through another family member or their friends ^[1].
3. Agree a safe phrase. The FBI recommends a secret word or phrase shared with family to verify identity ^[2]. Starling Bank launched its Safe Phrases campaign on 18 September 2024, in support of the UK government's Stop! Think Fraud campaign ^[8].
4. Limit public voice and video. The FBI advises limiting online content of your image or voice and making social media accounts private ^[2].
5. Refuse unusual payment methods. The FTC lists requests for wire transfers, cryptocurrency and gift cards as signs of a scam ^[1]; our advice is not to pay by these methods on a phone request.
6. Apply the same rule to any unexpected request. Since the FBI says criminals have used AI-generated audio to gain access to bank accounts ^[2], a voice alone is a weak proof of identity.

What does this look like in practice?

This is an illustrative scenario, not a real case. Your phone rings from an unfamiliar number. A voice that sounds like your daughter says she is in trouble and needs money now, by gift card. You ask for the family safe phrase. The caller dodges or pushes harder. You hang up, call your daughter on the number saved in your phone, and, if she does not answer, try her partner, a sibling or a friend ^[1]. You send nothing. Afterwards you report the attempt (see below).

Do safe phrases work? Strengths and limits

The strength is that a safe phrase does not depend on your ear and costs nothing. The limits are practical. It only works if the family agreed on it beforehand, everyone remembers it under stress, and it is not something a scammer could find online. The sources here do not measure how well safe phrases work, and the advice depends on you staying calm enough to use it. That is why the callback rule comes first.

What should you do if you already sent money?

Act quickly, but without blaming yourself. These calls are designed to overwhelm careful people.

1. Contact your bank or payment provider immediately. Tell them it was a scam and ask what can be done. Recovery is not guaranteed.
2. Report it. In the US, the FTC accepts scam reports through its ReportFraud service ^[1], and IC3 is the FBI centre that receives internet-crime complaints ^[3]. Elsewhere, use your national police or fraud-reporting body.
3. Warn your family. Tell the person who was impersonated and your relatives, since IC3 says the scam is evolving to mimic other family members and friends ^[3].
4. Review what audio of you is public. The FBI advises limiting online content of your voice and making accounts private ^[2].

This is general information, not legal or financial advice. For help with losses, contact a qualified adviser or the relevant local authority.

Are AI voice scam calls illegal?

Fraud is a crime, but the legal picture for AI voices specifically is still developing.

What did the FCC rule in 2024?

On 8 February 2024 the US Federal Communications Commission unanimously adopted a Declaratory Ruling that calls made with AI-generated voices are “artificial” under the Telephone Consumer Protection Act (TCPA), effective immediately ^[13]. This gives state attorneys general additional tools against AI-voice robocalls ^[13]. Law firm Cooley describes the ruling as a clarification of existing TCPA rules, not new rules ^[14]. Headlines said it “made” AI robocalls illegal; the law firm's reading is that it confirmed they already were.

The ruling is framed around a US law on calls made with artificial voices. It does not stop a criminal who phones a victim directly, and it has no effect outside the US.

What does the EU require?

The EPRS notes that Article 50 of the EU AI Act requires AI providers to inform users that they are dealing with AI-generated content, such as deepfakes, and to ensure it is detectable ^[6]. It lists technology, legislation and increased AI literacy and awareness as the routes to mitigation ^[6]. A labelling duty depends on the producer complying, which fraudsters are unlikely to do. For ordinary people, personal verification habits remain the main defence. This guide does not cover other countries' rules on voice cloning outside fraud.

Do scammers also use AI voices against businesses and officials?

Yes, according to FBI warnings. On 15 May 2025 the FBI said that since April 2025 malicious actors had impersonated senior US officials with text messages (smishing) and AI-generated voice messages (vishing) ^[12]. Its update of 19 December 2025 says the activity dates back to 2023, and that targets include officials' family members and personal acquaintances ^[7].

IC3's 2025 report also records AI involvement in business email compromise, with over $30 million in losses, in investment fraud, with over $632 million in complaints with an AI nexus, in confidence and romance scams, with over $19 million, and in employment scams, with almost $13 million ^[3]. Cloned voices are one tool among several.

The defence is the same as for families: do not act on a message or call alone. Verify through a second channel you already trust, such as a known number or a separate approval step.

What is still uncertain, and what is likely next?

What do we not know yet?

• True prevalence. Survey results such as Starling's 28% and complaint data such as IC3's $5 million in distress-scam losses are not comparable, and neither measures the full picture ^[8][3].
• How little audio is enough. The three-second figure comes from a bank's campaign; the FTC says only a short clip ^[8][1].
• Detection. The EPRS says detection tools have many limitations, especially for voice calls ^[6]. These sources contain no independent evidence that consumer-grade detection works reliably on phone calls.

What is likely next?

IC3 says the family emergency scam is evolving to mimic other family members or close friends in different emergencies ^[3]. Reported fraud overall is rising: IC3 losses were up 26% in 2025 and the FTC's total was about 25% higher ^[3][4]. Those totals do not isolate voice cloning, so they show the direction of fraud, not the size of this one scam. Laws such as the FCC ruling and the EU's Article 50 are in place, but their effect on fraudsters abroad is unproven.

What about readers in Iran?

The sources behind this guide are mainly from the US, UK and EU. None addresses conditions in Iran, so this article makes no claims about Iranian law, filtering or payment limits. The core steps, which are a known number, a second contact and an agreed phrase, depend only on you and your family, not on any particular network.