YouTube to begin automatically labeling AI videos

YouTube is increasing the visibility and automation of AI video labeling to help viewers distinguish real footage from AI-generated content. After earlier AI labeling that depended heavily on uploader disclosures, YouTube will require creators to indicate AI assistance during upload while also using new internal signals to flag AI content. The change targets videos showing significant photorealistic AI use, reflecting improvements in realism from models such as Seedance, Runway, and Google’s Veo. Google provides limited details about the signals, but identifies two firm triggers: C2PA metadata indicating a purely AI source and the use of watermarked Google tools like Veo. Creators can appeal incorrect tags, but labels tied to those triggers are permanent.

"Starting this month, YouTube will use “new internal signals” to flag AI content. This will apparently apply to videos that show “significant photorealistic AI use.” Google is vague about what signals will figure into its AI detection system—we’ve asked for more details and will update if we hear anything. The blog post does mention two ironclad triggers: C2PA metadata indicating a purely AI source and the use of watermarked Google tools like Veo."

"Creators are still required to indicate when uploading videos if they were created with the help of AI tools. However, uploaders didn't have any incentive to be honest about that before. Starting this month, YouTube will use “new internal signals” to flag AI content. This will apparently apply to videos that show “significant photorealistic AI use.”"

"When YouTube first attempted to tackle the identification of AI videos in 2024, it was almost gratuitous. AI videos at the time nearly always outed themselves by looking bizarre or disjointed. In just a few years, AI models like Seedance, Runway, and Google's own Veo have raised the bar for realism and consistency in AI video- the spaghetti is more accurate than ever."

"Creators who believe their videos have been tagged as AI incorrectly can appeal, but not if the site marks an upload as AI for either of those reasons. Those labels are “permanent.” Google is vague about what signals will figure into its AI detection system—we've asked for more details and will update if we hear anything."

Read at Ars Technica