What are deepfakes and how can you protect yourself?

Artificial intelligence is developing rapidly – and so are the methods for manipulating digital content. And of particular controversy are deepfakes. These are deceptively real videos, images or sound recordings that are created with the help of AI and deliberately falsified.

The consequences of deepfake can be serious – for example, if well-known personalities seem to make statements they never made. But deepfakes are no longer just about celebrities. Manipulated content can also become a threat in everyday life – for example in a private or professional environment. This makes it all the more important to understand how deepfakes work, what risks they entail, and how you can effectively protect yourself against them.

What are the dangers of deepfakes?

Deepfaking ranges from political misinformation to personal revenge campaigns to large-scale fraud and blackmail attempts. Deepfakes come from different threat scenarios, such as: 

• Social engineering: cybercriminals are very fond of using deepfakes for phishing attacks, also known as spear phishing in this area. They try to use the deepfakes to gain access to confidential information, such as bank or address data.  
• Defamation: false claims about alleged actions or words can seriously damage a reputation. Deepfakes make it possible to spread lies that massively damage the reputation of the victims. This also includes the mixing of pornographic content with the faces of non-participants. 
• Deepfakes in biometric systems: deepfakes can create content that looks and sounds like a real person. This is particularly problematic in the case of speech recognition or video identification methods. This is because the system cannot control where the recordings were made or whether they were tampered with. 
• Disinformation campaigns: some people deliberately use deepfakes for disinformation campaigns on the Internet. In the videos, they imitate well-known personalities who deliberately disseminate false information about a particular event. Experts often observe widespread disinformation campaigns on social networks in particular. 

Specific examples of deepfakes

Deepfakes have been repeatedly used to commit crimes in the past: 

• USD 35 million loss from voice swapping: through a fake phone call, criminals looted more than USD 35 million from a bank based in Dubai. Cybercriminals used deepfake voice-swapping and tricked the bank director of a major corporation that had a bank account there. From previous discussions, the bank director was familiar with the voice of his contact person in the company. So he did not doubt the authenticity of the call and approved a transaction of over USD 35 million.
• Deepfake cons the CEO of an energy company out of EUR 220,000: a CEO of a British energy company fell for a deepfake in 2019. His supposed supervisor of the parent company called him to request an emergency money transfer. The fake voice sounded so real that the CEO assumed he was talking to his boss. However, he was told not to transfer the money to the parent company, but to a third-party account. It was only when his boss asked for more money that the CEO's ears pricked up. However, it was already too late to repay the transferred amount.

How do you recognize deepfakes?

Cybercriminals are continually working on increasingly sophisticated scams. At the same time, cybersecurity companies are constantly developing new algorithms to detect such content. There is now a veritable race for technological advantage. But there are also tips for laypeople on how to expose manipulated content. 

Seven indications of a deepfake

There are a few signs for identifying a deepfake video: 

• Blank gaze: deepfakes can often be recognized by the person’s lack of facial expressions. Does the facial expression match the spoken message or does the face appear rigid and emotionless? 
• Unnatural eye movements: does the person blink at all or noticeably often? Are the eyes moving unnaturally?
• Blurred alignment: does the image have blurred edges or does the person appear to be positioned incorrectly in the image? Are individual frames distorted or unnatural? 
• Unusual body movements and proportions: is it noticeable that the person is jerky? Does the head appear large in relation to the body – or vice versa? 
• Unnatural teeth and hair: AI often has difficulty realistically depicting details such as individual teeth or flyaway hair. If such details are missing or appear false, this can be an indication of a deepfake. 
• Strange exposure or coloration: is the light unnatural? Does the lighting change noticeably from scene to scene? Does the skin color look strange or unnatural in front of the camera? Such features may indicate a video that has been tampered with. 
• Distinctive sound noises: are the lip movements consistent with what was said? Are there noticeable delays, pauses or even sound dropouts? Strange background noise can also be a sign of deepfake technology. 

How to check the authenticity of the content yourself

If you suspect a deepfake is behind a video, you should look at the source. As a first step, look at whether already known fact-check portals have examined the content. The most well-known portals include, for example:

• CORRECTIV
• dpa factchecking
• Mimikama

The most important thing is to remain skeptical. Of course, you don’t have to suspect a deepfake behind every video. A healthy dose of skepticism is appropriate for unrealistic or implausible scenarios. 

Are there any tools or technologies to automatically detect deepfakes?

In the fight against deepfakes, you are not completely on your own. There are now some tools that can help you detect deepfake videos, audios or images, such as: 

• Deepware scanner 
• FakeCatcher 
• DeepFake-o-meter
• Attestiv 
• Intel’s real-time deepfake detector 
• Video Authenticator Tool from Microsoft 
• WeVerify