A documented change of scale
In 2026, the FBI introduced a dedicated "AI-related" descriptor into its IC3 complaint center — an official acknowledgment that these scams now form a category of their own. In the very first year of this classification: more than 22,000 reports, for nearly $900 million in losses.
On the voice side specifically, Pindrop's (a voice-security specialist for call centers) 2025 Voice Intelligence and Security Report measured a 1,300% rise in voice deepfake fraud attempts in 2024, and estimates call centers' potential exposure at roughly $44.5 billion in fraud for 2025.
A documented case
In early 2024, an employee at a multinational company was deceived by a video conference entirely populated by AI-generated participants, impersonating several of the company's executives — to the point of authorizing a $25.6 million transfer. The case was documented by Hong Kong police and widely covered by specialist press.
Why it's different from a classic phone scam
A fake-bank-advisor or fake-tech-support scam relies on a script and the scammer's conviction. A scam powered by generative AI can additionally reproduce a recognizable voice — that of a relative, a boss, a child — from a very short audio sample, which makes the usual verification reflex ("I recognize their voice") far less reliable than it used to be.
Frequently asked questions
What is a voice deepfake?
A synthetic voice generated by artificial intelligence, trained to imitate a real person's voice — a relative, an executive, an advisor — from a short audio sample.
Does the FBI really track AI-related scams?
Yes. In 2026 the FBI introduced a dedicated "AI-related" descriptor in the complaint classification of its IC3 center: more than 22,000 reports and nearly $900 million in losses in the very first year.
Can you protect yourself from a voice deepfake?
The best reflex is still to verify through a second channel (call back a known number, send a message) before taking any urgent action requested by phone. Egidio also identifies known scam patterns, regardless of how realistic the voice sounds.