Deepfakes: Fraud using AI-generated voices is on the rise

The phone rings and the supposed grandson answers. He is in dire straits and needs money urgently. Older people in particular are tricked by this “grandchild trick.” The trick is old, but what's new is that the original voice is increasingly being used in the call, says Stefan Strauss of the Institute for Technology Evaluation at the Austrian Academy of Sciences (ÖAW): “Voice cloning refers to the use of artificial intelligence” to create a deceptive and authentic imitation of a snippet of real voice. “

Defines sound clearly

Using machine learning, the software learns from this snippet what the real thing sounds like. “Then you can speak any text with that simulated voice,” says the business IT specialist. The fact that this scam is becoming more common is partly due to software that can be used to create such counterfeits. Strauss says there are more and more of them and they are constantly improving. On the other hand, more and more audio recordings are being found online, which means there is enough source material for such scams.

When it comes to their fingerprints, most people are reluctant to reveal them online, Strauss says. However, when it comes to voice, only a few people realize that it is a biometric characteristic. The voice is so individual that it can be clearly assigned to a person. “With artificial intelligence, people's biometric characteristics can be recorded very easily and at the same time imitated, i.e. cloned. This requires more awareness and therefore more caution when publishing recordings of an individual's voice.

Help counter questions

The researcher advises basic skepticism when it comes to fake calls asking for money. The unknown number and money request should make you wary. “How reasonable is it for a relative to call and suddenly ask for money?” If you experience this, it is best to call someone else in the family and ask if that person has also heard about the emergency. Or ask the caller questions that only the presumed grandchild can answer. A family code word can also help.

As with any technology, there are not only bad but also good applications for sound reproduction, Strauss says: for example, when people can no longer speak due to illness and can only write to communicate. Then writing becomes a language again, even with your voice.