What is a Deep Fake?

Deep Fakes are deceptively real forgeries of video and audio recordings created using artificial intelligence. The word "Deep Fake" is a combination of "Deep Learning" and "Fake". Neural networks are used to replace or completely recreate faces, voices, or entire persons in existing recordings. Deep Fakes pose a growing challenge for IT security as they can be misused for manipulation and criminal purposes. By now, the entire internet is virtually flooded with deep fakes of celebrities and politicians.

How do Deep Fakes work?

The creation of Deep Fakes is based on machine learning, particularly on so-called Generative Adversarial Networks (GANs). Two neural networks are trained against each other: A generator creates forgeries, while a discriminator tries to distinguish these from real recordings. Through this competition, the forgeries become increasingly better. Large amounts of training material are needed for convincing results, such as video recordings or speech samples of the person to be forged.

Types of Deep Fakes

1. Face Swapping:
   This involves replacing one person's face in a video with another. This makes it possible to make people say or do things they never did.
2. Lip Syncing:
   With this technique, only the lip movements of a person are manipulated to make them appear to say something different. The rest of the video remains unchanged.
3. Voice Cloning:
   Here, a person's voice is synthetically replicated. The cloned voice can then speak any text that the real person never said.

What dangers do Deep Fakes pose?

Deep Fakes pose significant risks to individuals, companies, and society as a whole. They can be misused for various criminal and manipulative purposes. In general, there are three main dangers: disinformation, fraud, and blackmail. Deep Fakes enable criminals to make their attacks more credible and convincing.

Disinformation and Fake News

One of the biggest threats from Deep Fakes is the spread of false information. Fake videos of politicians or celebrities making controversial statements can spread rapidly on social media and influence public opinion. This can lead to political instability and undermine trust in institutions and media. Such Deep Fakes are often deployed strategically before elections or during crises to create confusion.

Fraud and Social Engineering

Criminals increasingly use Deep Fakes for sophisticated fraud schemes. One example is "CEO Fraud," where fraudsters imitate a supervisor's voice to prompt employees to make transfers. Identity theft is also facilitated by Deep Fakes. Cybercriminals can use fake video identifications for banking transactions or other sensitive operations. Therefore, it's important for companies to adapt their security measures and authentication processes.

Blackmail Attempts

If the attacker has access to many recordings of face and body, it's not difficult to create video material showing the victim in sexual acts. This type of video can be used to blackmail the victim. To prevent such a video from being sent to friends and family, one must pay money.

How can one protect against Deep Fakes?

Protection against Deep Fakes requires both technical and human measures. This should start with awareness and education. Users must learn to critically engage with media content and question sources. At the same time, researchers are working on technical solutions to detect Deep Fakes.

Technical Countermeasures

1. AI-based Detection Tools:
   Various algorithms can detect anomalies in videos and audios that indicate Deep Fakes. These tools are constantly being developed.
2. Digital Watermarks:
   Authentic media can be marked with invisible watermarks to prove their authenticity. This makes undetected manipulation more difficult.

Behavior-based Protection Measures

Besides technical solutions, heightened awareness of the issue is crucial. Users should critically question suspicious content and consult multiple sources. For important decisions or transactions based on media content, additional verification through other channels is recommended. Companies should regularly train their employees to minimize the risk of social engineering attacks using Deep Fakes.

Deep Fakes pose a serious threat to information security and trust in digital media. To address this challenge, a holistic approach is needed that encompasses technical innovations, education, and possibly legal regulations. Only in this way can a secure and trustworthy digital space be ensured in the long term.

Data Economy

With the rise of Deep Fakes, questions about data economy are becoming louder. Everyone should ask themselves whether it's necessary to upload videos and audios of themselves to the public internet, Facebook, Snapchat, or Instagram. Ultimately, this is training material for criminals. Until now, the availability of a certain relatively large amount of data has been necessary to create a Deep Fake.