Skip
CYBER TARGET
EN ▾

← Encyclopédie des arnaques

AI voice cloning scam : fonctionnement, signaux et que faire

Social engineering · alto

The scammer gets voice samples from social media videos, voice messages, interviews or even a short call where the victim says "hello, who's this?". AI tools generate new speech in that voice, live or as recorded clips.

Common uses: a call or WhatsApp voice note from a "child" in trouble asking for a transfer; a boss demanding urgent payment; a fake kidnapping with the victim's screams in the background; and ads with celebrity voices promoting fake investments or products. There are also attempts to fool bank systems that use voice as a password.

The FBI IC3 received 22,364 complaints with an AI component in 2025, with US$ 893.3 million in losses. Surfshark Research estimates global deepfake fraud losses of about US$ 2.5 billion in 2025.

Sinais / Signs

  • A familiar voice with an urgent request for money and secrecy
  • A call from an unknown or new number "because my phone broke"
  • Evasive answers to personal questions
  • Flat intonation, odd pauses or artificial background noise
  • A celebrity "recommending" an investment in an ad

Prevenção / Prevention

  • Agree on a code word with family and the finance team
  • Hang up and call the person back on their known number
  • Limit who can see your and your children's videos and audio online
  • Avoid talking much on calls from unknown numbers
  • Do not rely on voice as the only authentication factor at your bank

O que fazer / What to do

  • If you transferred money, alert your bank right away (in Brazil, request a MED claim for Pix)
  • Keep the audio, number and messages
  • Report to IC3 and the FTC (US) or file a police report (Brazil)
  • Tell the impersonated person so they can warn their contacts

FAQ

How much audio is needed to clone a voice?

Current tools can produce reasonable imitations from a few seconds, and quality improves with more material.

Is there an app that detects cloned voices?

Tools are in development, but none is foolproof. Verifying through another channel and using a code word remain the most reliable defenses.

FBI IC3 - Internet Crime Complaint Center · FTC - ReportFraud · FBI IC3 2025 Internet Crime Report (PDF, cópia) · Surfshark Research - Deepfake fraud statistics · CERT.br

EN · PT · ES · FR · ZH · AR