AI voice scams are making familiar voices unreliable

AI voice cloning has reached a point where scammers can convincingly imitate people, making a familiar voice unreliable proof of identity.

A 2026 paper, Evaluating AI Models' Capability to Automate Voice Phishing Attacks, by Fred Heiding and colleagues studied 4,100 US adults. Up to 36.1% said they would comply with, or were unsure about complying with, an AI-powered “relative in distress” scam. The figure measures reported willingness, not people who actually handed over money.

The researchers found that persuasiveness mattered more than how human a voice sounded. Regular AI users were no better at identifying synthetic voices than people without AI exposure.

Recent reports include a call in which an X user heard a voice resembling their wife, and a Reddit account describing callers who claimed to be an attorney and said a relative was in jail. It was unclear whether the first call used generative AI or a real-time voice filter.

A 2024 Ferrari deepfake attempt failed when an executive asked the caller, who appeared to be impersonating CEO Benedetto Vigna, about a book the real executive had recommended.

• Agree on a family code word.
• End unexpected calls asking for money.
• Call the person back using a known number.

A June Malwarebytes survey found that 85% said scams were hard to distinguish from genuine interactions, up from 66% the previous year. Another 88% said it was becoming harder to identify content that was genuinely human or real.