Voice cloning scam: How to spot fake AI calls mimicking your loved ones

Voice cloning scams that hinge on alibis like a distressed friend needing money or a cop demanding a bribe to clear a family member of an alleged offence, appear to be rising in India. What’s more? A study claims that 38 per cent of Indians can’t distinguish between the real and cloned voices of their loved ones.Â

Newspaper headlines about people losing thousands and lakhs in voice cloning scams have become a regular feature of our lives. The study cited above, conducted by McAfee last year, said that Indians are the most vulnerable group to such scams. As many as 83 per cent of Indian targets end up losing money, it claimed.

Online financial fraud accounted for 77.4 per cent of the cybercrimes from January 2020 to June 2023, according to a survey by the Future Crime Research Foundation (FCRF), an IIT Kanpur-incu­bated nonprofit.

AI-assisted voice scams are on the rise.

Voice clones & common fraud tactics

Voice cloning technology has advanced to the point where it only needs a few seconds of someone’s voice to accurately replicate it. It involves leveraging advancements in audio signal processing and neural network technologies to replicate a person’s voice.

With the fast evolution of AI models, it has now become easier for anyone to clone anyone’s voice with the help of AI-powered tools for free. All a person needs is a few minutes of voice data of the targeted person.

Publicly available digital persona of people make it easier for fraudsters to design social engineering campaigns for higher success rates.

“These campaigns are so contextualised that they make it harder for victims to distinguish between what is real and what is fake,” says Pradeep Yadlapati, who heads India operations of US-based IT firm Innova Solutions.

Pradeep Yadlapati says tech industry and policymakers need to work together for a robust policy framework to stop the misuse of artificial intelligence.

So, how do people avoid falling into these scams? More importantly, how can one determine if a voice is cloned or real?

Olga Yurkova, co-founder of dedicated fake content debunker site StopFake.Org, believes attentive listening is key to detecting fake audio. “If the person pretends to be your acquaintance, ask verification questions or verify through other mediums,” she told an online gathering of a group of journalists last week.

An ideal example is Delhi resident Gopal Garg. When he received a scam call earlier this year, the first thing he did was to verify the caller’s claims. “It was an early morning call. The caller identified himself as a cop from the Gurugram Police and said he had apprehended my son along with some other boys in a rape case. He demanded money to let my son go,” he said.

“They even tried to get me to speak to my son. Since I had already heard a lot about the ills of AI, I made it a point to first cross-check. When I called up my son, he said he was playing in a park,” Garg told indiatoday.in.

Gopal Garg emphasises awareness to protect people from voice scams.

Verification becomes even more important if the caller makes unexpected requests, particularly those involving urgent financial transactions or sensitive information.

If suspected scammers make a WhatsApp call – as has been the case with many victims in India, experts recommend disconnecting the call and dialling the same number for a normal mobile call.Â

Caller ID on smartphones could also be a great feature in such scenarios. Some experts recommend verbal codes for family and friends for emergency circumstances that will verify identity in those instances.

One of the tools employed in such scams is an effort to induce panic and emotional distress to cloud the target’s judgement. So staying calm and content is extremely important, says Yadlapati from Innova Solutions.Â

Advisory firm Ankura suggests people adjust privacy settings on social media for friends and family only to keep minimal exposure to public view online. It also recommends removing unknown or questionable individuals from friend listings.