YouTube boosts deepfake protections

YouTube has been tightening its defenses against AI-manipulated video through a mix of detection tools, disclosure rules and privacy-removal channels, even as social posts this week framed the moves as a fresh anti-deepfake push. The company has not published a single new May 19 announcement matching that description, but official YouTube pages show several related measures already in rollout or expansion. Those measures include a “likeness detection” tool for identifying AI-generated uses of a person’s face, mandatory disclosure rules for realistic altered content, and labels that YouTube can apply itself in some cases. ### What has YouTube actually changed? April and March updates from YouTube show the company widening access to its likeness-detection system beyond the first creator pilots. In a post published about four weeks ago, YouTube said it was expanding the technology to talent agencies, management companies and the celebrities they represent. In a separate post published two months ago, YouTube said it was also extending a pilot to government officials, journalists and political candidates. (support.google.com) September 2025 product updates had already said the company planned an open beta for all YouTube Partner Program creators. YouTube described the feature then as a way for creators to detect, manage and request removal of unauthorized videos made with their facial likeness. ### How does the deepfake-detection tool work? YouTube’s Help Center says “likeness detection” helps creators find content where their face appears to have been altered or generated by AI. (blog.youtube) If the system finds relevant content, enrolled users can review the videos and decide whether to seek removal through YouTube’s privacy complaint process. YouTube’s blog has compared the tool to Content ID. In the April post on entertainment-industry expansion, the company said the system looks for AI-generated content using a participant’s likeness, such as a deepfake of their face, and gives that person the ability to find it and request removal. (blog.youtube) The Help Center says the feature remains experimental, is not available in some countries, and only works for enrolled creators who consent and submit a facial reference. (support.google.com) It also says the tool cannot identify other people in videos unless they are enrolled. ### What are creators required to label? YouTube’s disclosure rules require creators to tell viewers when realistic content has been meaningfully altered or synthetically generated. (blog.youtube) The Help Center says disclosure is required when content makes a real person appear to say or do something they did not do, alters footage of a real event or place, or generates a realistic-looking scene that did not occur. (support.google.com) The company says those disclosures can appear in the expanded description or, in some cases, on the video player itself through a “How this content was made” label. YouTube also says it may carry forward disclosures using Content Credentials technology. ### Can YouTube add labels or remove content on its own? YouTube’s Help Center says the platform may proactively apply a label when altered or synthetic content is undisclosed and the company believes there is a risk of harm to viewers. (support.google.com) The same page says creators cannot remove that label in those cases. Privacy removals are handled through an existing complaint process rather than an automatic takedown rule for every deepfake. (support.google.com) YouTube says people who find unauthorized AI or altered uses of their image can ask the uploader to remove the content or file a privacy complaint with the platform. ### Why did this surface again this week? May 19 social posts appear to have bundled several ongoing YouTube AI-safety measures into one “deepfake protections” update. (support.google.com) The underlying official record points to a broader rollout rather than a single one-day policy launch: YouTube has been adding disclosure requirements, testing proactive labels, and expanding likeness detection to more groups over the past year. That reading is based on the timing of YouTube’s own blog and Help Center updates. (support.google.com) YouTube’s latest official guidance remains on its Help Center and blog pages, where creators can check eligibility for likeness detection, disclosure requirements for altered content, and the privacy complaint route for suspected deepfakes. (support.google.com 1) (support.google.com 2)