Scammers have increasingly used AI-based voice-imitation technologies to extort relatives of victims. The Washington Post reports.
According to the US Federal Trade Commission (FTC), in 2022 voice-based extortion involving impersonators was the second most common form in the country. During that period authorities received over 36,000 reports of people being deceived by scammers posing as friends and family.
According to the FTC report, more than 5,100 incidents occurred over the phone, causing more than $11 million in losses.
Advances in artificial intelligence have given scammers new tools for voice replication. A multitude of inexpensive AI-based online services can transform a short audio clip into a voice copy, enabling the scammer to make it ‘say’ any text.
Experts said regulators, law enforcement, and courts are ill-equipped to combat this kind of fraud. Also most victims have few leads to identify the perpetrator.
Law enforcement, in turn, finds it difficult to trace scammers’ calls, and courts have few legal precedents to hold the companies that develop such AI tools to account.
“It’s terrible. It’s a kind of perfect storm… with all the ingredients needed to create chaos,” said Hany Farid, professor of digital forensics at the University of California, Berkeley.
According to the expert, AI can analyse unique voice characteristics, including age, gender and accent. It can also locate similar speech in large databases and identify patterns.
These AI tools then recreate pitch, timbre and individual phonemes to produce a convincing overall likeness. According to Farid, a short audio sample taken from social networks and other public venues on the internet is sufficient:
“A year ago you needed a lot of audio to clone a person’s voice. … Now 30 seconds is enough.”
FTC Deputy Director of Marketing Will Maxon said tracking voice scammers can be “especially difficult”. They can use a phone located anywhere in the world. That makes jurisdiction even harder to determine for a given individual.
Maxon urged vigilance. He said that if you receive a call claiming a loved one is in trouble and needs money, you should verify the information.
Earlier in March, journalist Joseph Cox duped the bank’s voice ID using a free AI speech-synthesis service.
In January, the startup ElevenLabs was accused of disseminating racist audio deepfakes. Due to the rising number of abuses, the company was forced to suspend the service.