Deepfake videos mimic the face and voice of a person like a real person to deceive, request money transfer, steal identity.
What is Deepfake?
The term is a combination of the concepts “deep learning” and “fake”, referring to fake content created with artificial intelligence technology. Deepfakes are usually images, audio and video that masquerade a person’s face or voice for nefarious purposes.
According to the World Economic Forum (WEF), deepfake videos are growing at a rate of 900% annually. Recent technological advancements have made it easier to manufacture them.
Deepfakes are used as part of an attack to influence operations or launch disinformation campaigns. Other uses include online fraud, identity theft, and financial fraud.
One of the most popular deepfake examples in recent days is the image of former US President Donald Trump being arrested.
In terms of working mechanism, deepfake uses advanced artificial intelligence techniques that collect data on physical movements, facial features and even voice. It process it through AI or Network encryption algorithms. Generative antagonist (GAN) to create fake but extremely realistic sounds and images.
In other words, it’s an AI-generated video, image, or sound that mimics a person’s appearance and/or voice.
This is not entirely new technology. In fact, it has been used for many years in Hollywood studios, but is now gaining popularity through commercial applications. This content increased so much that Facebook banned deepfakes in 2020.
The Dark Side of AI
Deepfakes is not bad, but are often used for shady purposes. A recent Europol report warns that most spreading deepfakes have malicious intent. Things where this technology is often exploited include: Match people’s faces into context with their own intentions. Use a person’s picture and voice to bypass biometric passwords. Fraud on digital platforms. Spreading fake news and misinformation risks disrupting financial markets and destabilizing international relations. Identity theft, extortion.
As this technology becomes more and more accessible, the number of crimes could increase. Therefore, the authorities advise the public to have a clear understanding of deepfakes and how to prevent the associated threat.
There have been many deepfake scams recently, such as crooks using deepfake technology to create holograms, then using them in video calls to pose as communications directors, fooling other executives to collect confidential information.
With deepfake sound, the threat actors used real-time voice transcription technology to trick a Hong Kong banker into transferring $35 million.
In one particularly egregious scam, the threat actors sent voicemails simulating a CEO’s voice, asking employees to donate to charities and disaster relief through fake websites. to transfer money to accounts abroad.
How to detect deepfake?
Deepfakes are becoming more sophisticated and harder to detect. But there are still some ways to detect if what’s in front of you is a scam or not.
Number of eye blinks
Paying attention to the number of times the image in the video blinks, we can tell if it’s a real person or a deepfake, because deepfakes tend to do so less than real people, sometimes in a forced or unnatural way.
Face and body
Imitating a person’s entire personality requires quite a bit of work, so most deepfakes are limited to replacing faces. One way to detect spoofing is to identify mismatches between body-to-face proportions, facial expressions and body movements or postures.
Video length
A fake video requires several hours of execution and algorithm training, so deepfake videos are usually only a few seconds long.
Audio and video
Deepfake is still limited in combining both voice and face fake. Be suspicious if the video has no sound or has audio that doesn’t match the aperture.
Inside the mouth
Deepfakes are not good at faithfully reproducing the person’s tongue, teeth, and oral cavity as the person speaks. Therefore, the blur inside the mouth shows that it is a fake image.
Other details
Details are also the weak point of deepfake software. Therefore, we can detect them by focusing on small aspects, such as dark circles around the eyes, unreal facial beard, overly smooth or wrinkled skin, false moles and lip color that is not nature.
