Managing your facial likeness with likeness detection (support.google.com)

YouTube has launched an experimental, opt-in “Likeness detection” tool that scans new uploads to find videos where a creator’s face or voice may have been altered or synthetically generated. Creators enroll through YouTube Studio by submitting a brief video selfie and government ID for identity verification; the system creates face and voice templates (derived from that selfie and your YouTube content) and compares them to new uploads. Matches are surfaced for review so creators can flag AI-altered content and submit privacy removal requests, or mark clips as real footage, not their likeness, or archive them. The feature is limited to select countries and eligible creators who consent, and YouTube says it will not identify people who haven’t opted in. This matters because it gives creators an automated way to detect deepfakes and manipulated likenesses at scale—similar in concept to Content ID but focused on biometric matches—potentially speeding takedowns and abuse prevention. Key technical and policy details: biometric templates are stored up to three years (or until consent is withdrawn), a screenshot from your verification video may be used in review, and you can opt to allow templates to help train detection models (revocable). YouTube cautions some matches may be real footage (not removable under privacy rules) and emphasizes consent, limited scope, data retention, and that the tool is experimental, raising important questions about privacy, verification security, false positives, and how biometric data is handled.