Sentinel, a startup developing a synthesized media or deepfakes identification platform, recently raised $1.35 million in a seed funding round. The major investors were Jaan Tallinn from Skype, Taavet Hinrikus from TransferWise, Ragnar Sass, Martin Henk from Pipedrive, and United Angels VC.
Sentinel works on a multi-layered defense approach as it is a challenging task to detect all deepfakes with only one detection method. It offers four layers of deepfake defense. The initial layer uses known examples of deepfakes across social media platforms to check against any such deepfakes. The second layer consists of a machine learning model that parses metadata for manipulation. The next layer checks for audio changes, looking for synthesized voices, etc. Finally, a technology analyzes faces “frame by frame” to look for visual manipulation signs. After going through a rigorous detection phase, the output is finalized from every layer’s result, and an overall score is given that determines whether it is deepfake or not.
The company has a vast database of in-the-wild deepfakes to train its algorithms. It collects data from all the major social media platforms like YouTube, Facebook, Instagram, TikTok, etc., to generalize it and trains its model using this data. But even then, it cannot detect deepfakes with 100 percent accuracy.
Sentinel’s primary target is to help governments and media organizations minimize risk and protect against deepfakes and other malicious information ops through their automated digital media authentication platform. Their technology is being used by leading organizations, including European Union External Action Service and the Estonian government.
According to the startup, deepfakes are growing exponentially across all social media platforms. Tools to create deepfakes are getting more accessible day by day. While plenty of them are harmless, but many may cause serious problems. To scale up the deepfake detection technology is the need of the hour. There is a cut to cut competition in this field, with tech giants like Microsoft, which earlier this month launched a detector tool to help pick up disinformation aimed at November’s U.S. election.