On March 12, 2024, the Federal Office for Information Security (BSI) published guidance on deepfake dangers and countermeasures. The guidance includes information on the falsification of videos, texts, possible threat scenarios, countermeasures, and challenges of automated countermeasures. 

What are the suggested countermeasures?

The BSI stated that there are various approaches to defend against deepfakes, which can be divided into two categories, namely prevention and detection.

For preventive measures, the BSI notes, among other things: 

• Knowledge and training: a key measure against deepfake attacks is the training of potentially affected persons. Many deepfake processes sometimes produce clear artifacts and knowledge of these artifacts can significantly increase the detection of forgeries.
• Cryptographic methods: offer the possibility of clearly linking the source of the material to an identity. However, this cannot prevent the source itself from tampering with the material beforehand. Current developments, for example, focus on creating a digital signature during the recording process, which ensures that the material has not been tampered with after it has been recorded.
• Legal regulations.

For detection, the BSI notes, among other things: 

• Media forensics: it is possible to detect artifacts that arise when manipulation methods are used.
• Automated detection of manipulated data: these methods are usually based on techniques from the field of artificial intelligence (AI), in particular deep neural networks. However, it should be noted that automated detection methods often only work reliably under certain conditions. Since these methods are usually based on AI methods, they come with their own fundamental problems.

You can read the guidance, only available in German, here.