YouTube used its Made on YouTube event to say that later this year its likeness detection system, which scans uploads for AI deepfakes of enrolled creators' faces, will also check their speaking voices. The company says pairing the two signals should make matches more accurate when someone's likeness turns up in deepfake content, and prepare the ground for wider voice protections. It framed the change as part of a broader package of creator tools, arguing that protecting a creator's face, voice and personal brand matters more as the platform grows.
The face tool began as a small pilot and was widened to politicians, officials and journalists, then the entertainment industry, and finally all eligible creators over 18, who enrol with a government ID and a selfie video. That builds a reference the system uses to search new uploads for possible matches.
YouTube also plans to bring enrolment, match alerts and takedown requests to its mobile app.
Creators who enrol can separately choose whether their face and voice templates may be used to train YouTube's detection models, and can withdraw that consent later. Biometric Update notes that a 2025 Resemble AI report put YouTube at nearly 30 percent of documented deepfake incidents, making it the most used platform for spreading them. It also cites an Avast survey in which 56 percent of adults said they were sure they would spot an AI fake on sight, even as synthetic video and audio become harder to tell apart from the real thing.
Sources · 1