Artificial intelligence fuels voice impersonation scams in the US

A new wave of sophisticated scams is exploiting artificial intelligence to impersonate individuals, leaving victims in the United States facing significant financial losses. Liz Benz, a 46-year-old insurance broker, recently experienced this firsthand when she received a distressing call from what she believed was her 16-year-old son. The voice, eerily similar in tone and cadence to her son's, was actually an AI-generated clone.

“

Nothing would have convinced me it was a scam until I saw it with my own eyes. It was 20 minutes of panic.

The scammer, using the cloned voice, claimed Benz's son was in trouble after a friend was shot and killed, and that the son was being held hostage. The demand was immediate: deliver cash to a nearby supermarket to secure his release. Benz was only convinced it was a scam when she received a smiling selfie of her son from the soccer game, realizing she had been targeted during a harrowing 20-minute ordeal.

“

Before it was difficult to do. Now anyone can do it in seconds.

Authorities and consumer advocates are increasingly warning about these voice-cloning scams, a modern twist on the classic 'uncle scam.' The FBI revealed that Americans lost over $893 million to AI-enabled fraud in the past year alone. The technology is now widely accessible, with numerous voice-cloning applications available online, many for free, capable of producing convincing replicas from short audio clips.

“

A guy alone in a room with a keyboard can create an infinite number of scammers.

Brian Long, CEO of Adaptive Security, noted the alarming ease with which these scams can now be executed. "Anyone can do it in seconds," he stated, explaining that a single individual with a keyboard can generate countless fraudulent calls. These AI tools can utilize audio from social media or voicemails to create convincing impersonations. Often, the goal is not a perfect replica but to create enough urgency and doubt to prompt immediate action without verification, according to Amit Gupta, an executive at cybersecurity firm Pindrop.

“

A distressed voice saying 'Mom, help me' or 'Dad, I had an accident' only has to sound credible for a few seconds.