The popular short video platform TikTok has begun testing a new tool called Likeness Detection. Currently, the feature is available to a limited group of creators in the US. The essence of the mechanism is automated search for content created using artificial intelligence that copies a specific blogger's appearance.
Once the system detects a suspicious deepfake or fake account, the user gets the ability to promptly file a complaint about the violation. However, to activate this protection, users must undergo strict identity verification through a third-party service Jumio. The procedure includes a real-time selfie and scanning of an identification document.
An important point: TikTok states that it does not save copies of uploaded documents. Biometric data is used solely to create a unique template, which is then used to search for matches in videos. This is standard practice for privacy protection, but it raises questions about the level of trust in centralized biometric systems.
Expert Opinion
From a technical standpoint, this is a step in the right direction. However, the effectiveness of Likeness Detection will directly depend on the quality of the AI model, which must be able to distinguish a real face from a generated one. Given the rapid development of generative neural networks, the battle between deepfake detectors and generators is just beginning. For now, this testing is more of a marketing gesture than full-fledged protection, but it's a start.