Shock calls have long been a favoured tactic to obtain sensitive data or money from unsuspecting victims. The scam is now largely recognised and many citizens are already on alert. Someone calls, claiming that a family member, typically one’s child, has been involved in an accident. They suggest that only money can resolve the situation. At this point, most people recognise the scam and end the call. But what if the voice on the other end isn’t a stranger, but seemingly your own child? The situation then becomes far more concerning, and the natural instincts of worry and willingness to help kick in. With the aid of deep fakes, fraudsters elevate the shock call scam to an entirely new level. Thomas Wrobel, spam protection expert from Clever Dialer, discusses how criminals exploit the imitation of familiar voices and what individuals can do to protect themselves against this new type of fake call.

AI-generated portrait of a woman
Is this person real or has AI played a part? | Source: Unsplash

AI: Artificial Intelligence or Criminal Intent?

Since the wider adoption of tools like ChatGPT, artificial intelligence (AI) and deep fakes have become hot topics of discussion. Beyond their potential, it’s primarily the criminal risks posed by this new technology that’s causing concern among the public. Lately, there have been increasing reports of citizens being harassed using the age-old shock call scam. However, the new twist is that these callers use deep fakes to mimic the voices of real family members. The term “deep fake” combines “deep learning” and “fake”. The former refers to machine learning that analyses data and recognises patterns, allowing artificial content, including voices, to be created, manipulated, and thus faked. Voice imitation technology has advanced to the point where it can almost flawlessly mimic others.

A recent case from Saarland illustrates the audacity and ruthlessness of these scammers. A 54-year-old woman received a call from someone impersonating her “daughter”, claiming to have been involved in a fatal road accident. The conversation was quickly taken over by another individual posing as a police officer, demanding €80,000 bail for the woman’s daughter. Although the distraught mother was ready to withdraw the demanded amount from her local bank, a vigilant bank employee managed to prevent the scam. Not just in Germany, but shock calls leveraging new AI capabilities are on the rise globally. In the US alone, an estimated 5,000 individuals were scammed this way last year, with damages amounting to around $11 million. But how do these crooks access genuine voices?

The Perfect Copy Thanks to Social Media

Social media platforms have long been a double-edged sword. In terms of fraud, they’re primarily a valuable resource for obtaining voices of potential targets. Recordings from TikTok, Instagram, and similar platforms are utilised to train the synthetic voice. This way, AI learns to sound like real individuals and imitate genuine family members. Language barriers are no longer an obstacle. In the past, a scam could often be spotted if the caller had a broken accent. However, programmes like ChatGPT can effortlessly and instantly produce flawless texts that just need to be read out by the faked voice. This means grammar is no longer a hurdle for criminals, and the threat from fake voices continues to grow. Especially considering that successful scams used to require winning the victim’s trust, fraudsters now have an easier time as the mimicked voice is usually already trusted. The risk of quickly losing one’s savings has thus multiplied.

Listening Carefully and Mentally Going Through Improbable Scenarios

Despite these menacing advancements, citizens shouldn’t forget there are ways to guard against the escalated shock calls. For instance, families and close friends can set up a password to unmistakably identify the caller. Particularly when exorbitant sums are demanded or unusual scenarios are described, asking for the “password” can provide clarity. Furthermore, the flow of the conversation should be scrutinised. The technology isn’t yet perfect for maintaining prolonged dialogues without noticeable quirks in pronunciation. Hence, the supposed family member often hands over the phone to another person. If there’s a consistent voice on the line, listen closely: various background noises can produce a metallic sound in the voice. If your daughter sounds like a robot, it’s probably one. Proper nouns, specific – mostly English – terms, or dialects can’t yet be perfectly mimicked.

If in doubt, end the call and dial the trusted contact directly. It’s crucial not to use the call-back function but to manually input the number, as the displayed number might have been masked (ID-Spoofing). Above all, common sense should prevail. As terrifying as a shock call might be, one should always remember that in a genuine accident, no monetary demand would be made for exoneration or medical treatment. Neither doctors, nor lawyers, nor the police would prioritise a financial request over the welfare of the injured. Hence, mentally walking through such emergency scenarios can be helpful, preparing oneself for any purported dire situations.