Thieves are using AI’s increasing skill at cloning people’s voices to steal money from bank accounts.

In one instance cited by The New York Times, a man phoned his banker to discuss a major funds transfer he was planning.

When the banker took the customer’s next call, it wasn’t the customer at all. Instead, it was a software program that had cloned the customer’s voice and tried to persuade the banker to send the money to a different account than the one the customer had indicated earlier.

Scammers need only a small sample of a voice from a TikTok or other social media post and today’s AI programs can take it from there.

Microsoft’s VALL-E program needs only a three-second speech sample to recreate a person’s speech patterns and pronunciations, Vijay Balasubramaniyan, president of the voice-verification service Pinpoint, told the NYT.

“Imposter calls” make up a tiny fraction of the thousands of fraudulent calls banks’ customer service centers receive each year. The real danger lies in scammers calling individuals, he said.

Increasingly sophisticated synthesized speech avoids the artifacts that security algorithms lock onto, such as long pauses or the sound of keystrokes, and can more and more easily fool even an astute person.

As an example, security consultant Rachel Tobac appeared on a “60 Minutes” episode in May and used software and five minutes of programming time to clone correspondent Sharyn Alfonsi’s voice so convincingly that Tobac was able to coax a CBS employee to give her Alfonsi’s passport number.

With that piece of data, a skilled identity thief could rampage through a person’s financial life.

TRENDPOST: The battle between AI-equipped thieves and scammers and security algorithms will not end; each will learn from the other in an increasingly sophisticated contest.

In the age of AI, machine learning, and social media, the only way to ensure security in financial transactions is to do them the old-fashioned way—using paper checks or going face to face with another human.