DeepFake Check
Voltar ao Blog
DeepFakeCheck Team 7 min read

AI Voice Scam Protection: How to Spot Cloned Voices Before They Cost You

TL;DR — The 4 Rules That Stop a Voice-Clone Scam

If you only remember four things from this guide:

  • 1. Hang up. Call back on a number you already know. This single habit defeats 95% of voice-clone scams.
  • 2. Establish a family safe word. A code phrase that only family members know. The scammer cannot fake what they do not know.
  • 3. A real emergency from a real person can wait three minutes. Anyone pressuring you to act in 30 seconds is running the script.
  • 4. If you have a recording, run it through an AI voice detector. A confidence score is a useful sanity check, but the rules above are your primary defense.

The rest of this guide explains why voice-clone scams work, the specific audio signs to listen for, and what to do after you spot one.

Why voice-clone scams exploded in 2025–2026

Three things changed at once:

  • Voice cloning got cheap and fast. As of 2026, services like ElevenLabs let anyone clone a voice from a 30-second sample for less than $20/month. The sample can come from a public Instagram reel, a podcast appearance, or a leaked voicemail.
  • Phone networks still don't authenticate caller ID. STIR/SHAKEN deployment helped, but spoofing a familiar number remains trivially possible in most countries.
  • Targeted attacks became economical. AI lets scammers personalize each call with the victim's name, the relative's name, and a believable scenario, scaling what used to be artisanal fraud.

The result: the FBI's IC3 reported a 350%+ jump in voice-clone-related complaints between 2023 and 2025. The pattern is depressingly consistent.

How the scam runs (the script)

The classic family-emergency variant goes like this:

  • 1. The call comes from an unknown or spoofed number — often masked to look like a hospital, jail, or out-of-area number.
  • 2. A panicked voice that sounds like your relative says they've been in an accident, arrested, or kidnapped.
  • 3. A second "official" voice (a fake lawyer, a fake officer, a fake doctor) takes over and demands money — usually via wire transfer, gift cards, or cryptocurrency.
  • 4. You are told not to hang up or call anyone else. This is the critical pressure tactic.
  • 5. You are pushed to act within minutes. Urgency is the entire scam.

Business-targeted variants substitute the relative for a CEO authorizing an emergency wire transfer ("CEO fraud"). The script is otherwise identical.

The audio signs of a cloned voice

Even high-quality voice clones leave traces. Listen for:

1. No breath noise

Real speech contains audible inhalation between sentences. AI-cloned voices have unnaturally clean silences. If the voice sounds "studio-quality" coming from a relative's mobile phone, that's suspicious.

2. Identical sentence rhythm

Real speech varies in cadence, especially under stress. AI clones tend to produce sentences with eerily consistent pacing — the model can't easily reproduce the broken, jagged speech of someone actually panicking.

3. Over-articulation

Voice clones often pronounce every consonant cleanly because the model was trained on read audiobook samples. Real distressed speech has slurred consonants, half-finished words, and trailing endings.

4. Background that doesn't match the story

The scammer's room background often has subtle clues — air conditioning hum, traffic, distant voices — that don't match the claimed location. A "relative in a jail cell" with a quiet, anechoic background is a giveaway.

5. The voice doesn't react

Real people respond to specific questions. Voice clones, especially live ones, often loop back to the script when you ask a question they don't expect. Try asking "What's our dog's name?" or "Where did we have dinner last Tuesday?" — a clone will deflect.

The verification routines that actually work

The hang-up-and-call-back rule

This is the single most effective defense. The moment you have any doubt:

  • Say "Let me call you right back" and hang up.
  • Look up the relative's number from your contacts (not from the screen of the current call).
  • Call them. If they answer normally, you just dodged a scam.
  • If they don't answer, call another family member to confirm before doing anything.

The scammer relies on emotional momentum. Breaking the call breaks the script.

The family safe word

Agree on a phrase that only family members know — not your dog's name (those are often on social media), but something arbitrary like "purple kite" or "Tuesday's lemon". When something feels off, ask for the safe word. A real relative who's actually in trouble will not be offended; a scammer will fumble or hang up.

Update the safe word every six months. Keep it out of group chats and email.

The "wait three minutes" rule

No legitimate emergency requires money in under three minutes:

  • Hospitals do not demand pre-payment via wire transfer or gift cards.
  • Police never ask for bail in cryptocurrency.
  • Real lawyers do not call random family members directly.
  • Real kidnappers, in the rare event they exist, give you hours, not seconds.

If you are told you have 30 seconds, the urgency itself is the proof of fraud.

Use an AI voice detector for recordings

If the suspicious audio is recorded — voicemail, a forwarded clip, a video — run it through a detector. DeepFakeCheck's voice detector is free and gives a confidence score in seconds. We support all major voice generators (ElevenLabs, OpenAI TTS, Murf, Resemble) and the result includes audible artifact indicators that explain why the audio is suspicious.

For live calls, none of the consumer-grade detectors can analyze in real time. The behavioral rules above are your defense.

What to do if you've already been scammed

Time matters. In order:

  • 1. Stop sending money. Even if you've already sent some, do not send more.
  • 2. Call your bank. Wire transfers can sometimes be reversed if reported within 24 hours. Cryptocurrency transactions usually cannot.
  • 3. File with law enforcement. In the US: FBI IC3. In the UK: Action Fraud. In Germany: BSI or local police. In Korea: 보이스피싱 신고센터 (Cyber 112). Most countries now treat voice-clone fraud as a priority crime.
  • 4. Tell your family. Not just to warn them — they may have received the same scam call. Pattern matching helps law enforcement.
  • 5. Save everything. Phone records, audio recordings, screenshots of transfer confirmations. Useful both for criminal complaints and for insurance.

Common questions

How long does the scammer need of my voice to clone it? As little as 3 seconds for low-quality clones, 30 seconds for convincing ones, and 5+ minutes for clones that can fool a family member. Public podcasts, YouTube videos, Instagram reels, and voicemail greetings are all viable sources.

Can my bank tell it's a scam call? Sometimes. Many banks now train staff to recognize the emotional state of voice-scam victims and to delay or block urgent wire transfers. Call your bank's fraud line if you've been pressured — they would rather pause a legitimate transfer than process a fraudulent one.

Is my elderly parent at higher risk? Yes, statistically. But the highest-loss attacks in 2025–2026 actually targeted professionals — finance team members tricked by CEO impersonation. The voice-clone scam is age-agnostic; the variant differs.

Does anti-spam phone software help? Modern apps (Hiya, Truecaller, the iOS/Android built-in screening) catch obvious scam patterns but cannot detect a voice clone in real time. Treat them as one layer, not the whole defense.

The honest bottom line

The technology to clone a voice is already cheap. The technology to detect a clone in real time during a phone call is not yet available to consumers. That gap is exactly the scam's business model.

What stops voice-clone scams is routine, not technology: hang up and call back, use a safe word, refuse to be rushed. These habits cost nothing and break every script we've seen in production data.

If you have an audio file you want to verify, try the free voice detector at DeepFakeCheck. For everything else, trust the routine.

Acha que essa imagem foi feita por IA?

Use nossa ferramenta de detecção de deepfakes de alta precisão para analisar imagens na hora.

Analisar imagem agora