Identifying impersonating deepfake calls

One night in late April, Tuyet Thu (Ho Chi Minh City) continuously received more than a dozen calls from friends and relatives asking why they needed money urgently. "I was on vacation, I didn't need to borrow money, but everyone confirmed that they had talked to me directly on a Facebook video call," Thu said.

A day earlier, she had lost her Facebook account and was still trying to appeal to regain access. Thu had to ask her friends to inform her that she had been hacked so they could be on the lookout and avoid being scammed.

Thu is not the only one who has been in trouble because of deepfake videos. In a survey conducted by reporters on April 25, 8% of respondents said they had fallen for a deepfake call. 20% said they had received one but were not fooled.

Deepfake is a combination of deep learning and fake. This technology uses AI to analyze a person's gestures, facial expressions, and voice, then recreates and edits them to create photos or videos that look real.

"Bad actors can create a fake video with just a photo and 5 seconds of the victim's voice. The more photos and videos users post online, the easier it is for cybercriminals to create deepfakes," said Dr. Ngo Tan Vu Khanh, a security consultant at Kaspersky.

Deepfakes will become more sophisticated, making it difficult to distinguish between real and fake, but there are still some limitations for users to recognize.

Signs of deepfake video

Deepfakes of images and voices are often used by bad guys to gain the trust of victims. However, these videos have in common low image quality, blurriness, and short duration. In addition, users can also pay attention to some signs such as:

Unnatural eye movements: Current technology cannot mimic the natural way humans blink. When communicating, eyes often change according to the movements of the other person, but deepfake videos cannot do that.

Unusual skin tone and lighting: If you notice the caller’s skin tone changing from frame to frame, it could be a deepfake. If in doubt, look for unusual light, shadows, and streaks of color across the frame.

Weird movements: Deepfake calls are often of poor quality, lacking smoothness, easily revealing traces when turning the head, leaning to one side, and the frame is jerky.

Asymmetrical face: If you see a face that doesn't show much emotion, the details of the eyes and nose are not similar, each facing to one side, it could also be a deepfake video.

Mouth Shapes: Deepfakes often combine video and audio from two separate files, so mouth shapes don't match. Additionally, voices from robots or software often have digital noise and lack natural environmental sounds.

Duration: Deepfake calls are usually very quick. The attacker usually calls for 5-7 seconds and then hangs up, claiming the signal is weak or they have run out of money. When the victim calls back, they often can't get through or just say the same old things.

According to Mr. Vu Khanh, the purpose of deepfake calls is financial fraud, stealing information from users. When receiving calls related to sensitive information or issues, users need to be alert, check and call the relevant parties to confirm, avoid falling into traps.

Avoid falling victim to deepfake

According to Kaspersky security experts, there have been some studies using AI to detect deepfakes in real time. However, these tools are not yet popular.

The best way to avoid becoming a deepfake victim is to proactively use software to add digital noise to photos and videos before posting them online. This has almost no impact on quality because it is difficult to detect with the naked eye, but it takes time and effort.

Additionally, users should protect themselves by limiting their public presence on social media. Enable privacy settings to prevent scammers from easily stealing their voice and likeness.

According to Mr. Khanh, users should also get into the habit of backing up data to easily restore it if their account is stolen, using strong and different passwords for each account, regularly updating software, and always being vigilant when providing sensitive information, especially related to finances, on the Internet.

According to VnExpress