Warning system against deepfakes: Tom Cruise in the teams switch

When Martin Steinebach presents a video call with his deepfake detector to interested companies, he is often met with disbelief. Because the man he is talking to is known to everyone present through his numerous action films. It's the actor Tom Cruise.

Admittedly, the US star's head movements are a bit jerky here and there, but that's common in video conferences. However, it is surprising that the Mission Impossible hero speaks fluent German. Or not.

Because the man sitting in front of the laptop camera is not Tom Cruise, but Steinebach's employee Raphael Frick. The doctoral student played a key role in developing deepfake detection in real time. And not primarily for show purposes.

In order for the warning system, which was developed by Martin Steinebach's team at the Fraunhofer Institute for Secure Information Technology (SIT) in Darmstadt, to react as efficiently as possible, it must have the ability to distinguish between real and fake, between real people and deepfake - and in almost real time. This is the only way to report suspected fraud to the potential victim during the conference.

Just as the cybercriminals' deepfake programs are AI-generated, the warning program developed by the researchers also uses artificial intelligence (AI). The AI ​​analyzes millions of data from the fake programs or deepfakes created at the institute itself using AI - and learns independently how real video images differ from fake ones.

The effort that researchers put into developing such warning systems shows how great the danger has now become - and the enormous economic damage that can occur if warning mechanisms are missing.

“There are enough real cases of fraud these days,” says Steinebach. For example, a case from Hong Kong became known worldwide in which criminals managed to impersonate the entire board of directors in a video conference with an employee and get the employee to transfer 25 million US dollars to a third-party account.

Real-time deepfakes in video calls represent one of the most dangerous new levels of cybercrime worldwide. Steinebach has also noticed an increased awareness of this in this country. According to studies, identity theft and identity imitation are one of the biggest challenges for large companies and not least banks, says the expert. Because it opens up “a completely new fraud vector that has so far been difficult to deal with.”

The deepfake warning program is still in the internal testing phase. They will soon be working with companies and providers of video conferencing systems such as Microsoft teams or Zoom to launch the system in a user- and data protection-friendly manner - with the aim of making deepfake fraud a real mission impossible for cybercriminals.