Researchers bypass the security of Telenet’s sexting app with AI

Researchers from Ghent University have developed a tool that uses artificial intelligence to undo the watermark protection of Telenet’s .comdom app. The researchers made the tool in a few days.

The researchers regret that Telenet paid insufficient attention to security when developing the .comdom app. The provider released its .comdom app at the end of November to protect the mutual sending of spicy photos. The app puts a watermark with the recipient’s information such as name and phone number over the image and makes the sender’s face unrecognizable. “This filter cannot be removed even by the experienced Photoshop user – or at least not without making the nude photo unrecognizable,” Telenet wrote about its app. The app should prevent recipients from further spreading the spicy photos, because they can be traced back to them.

However, researchers Martijn Courteaux and Hannes Mareen of the IDLab-Media research team at UGent managed to automatically undo the watermarks in a short time. According to the creators, their software can “nearly reconstruct” the original photo, removing the recipient’s data. The researchers did this by finding out exactly how the watermark is built up by the app, and by providing thousands of photos with such a watermark themselves. They then trained a neural network on that collection of photos and the original variants. The trained algorithm was then able to recognize watermarks and construct a photo as it would have been the original.

According to the researchers, the properties of the watermarks were too simple and the power of AI algorithms was not taken into account during development. They also note that the detection of the app to make faces unrecognizable does not always work well and also does not function with multiple faces.