YouTube says it will start automatically detecting and labeling videos with “significant photorealistic AI-generated content,” shifting away from the creator self-disclosure system it introduced in 2024. The automated labels begin rolling out gradually in May 2026, using internal signals YouTube has not fully specified.
For anyone who has ever gone spelunking in a video description: the labels are moving into daylight — below the player for long-form videos and as an overlay for Shorts.
Creators can still contest an automated flag, but YouTube says labels will be permanent in two cases: videos made with its own AI tools, including Veo, Gemini Omni, and Dream Screen; and files with C2PA metadata showing the content is fully AI-generated.
C2PA, the Coalition for Content Provenance and Authenticity, is an open provenance standard founded in 2021 by Adobe, Arm, the BBC, Intel, Microsoft, and Truepic. The article says it has grown to more than 6,000 member organisations. YouTube’s system will also be able to read Google’s SynthID imperceptible watermarks, which the piece says have been applied to more than 100 billion AI-generated images and videos to date.
The label expansion sits next to broader deepfake rules. As of 16 May 2026, any adult aged 18 or over can request removal of face-based deepfakes depicting their likeness; previously, the option was limited to public figures, including some creators, celebrities, politicians, and journalists. Voice cloning detection is expected later in 2026.
YouTube says AI labels will not affect recommendations or monetization. They are meant to be informational, not punitive. The timing is still hard to miss: EU AI Act transparency obligations take effect in August 2026, though YouTube has not framed the change as compliance-driven.
There is also the product-roadmap awkwardness. YouTube is adding AI features — Ask YouTube, AI playlist generation, AI video summaries, Gemini Omni in Shorts Remix and YouTube Create — while also making AI outputs harder to pass off as ordinary uploads. The platform is choosing labels over restriction. Whether viewers treat those labels as useful context or as another cookie-banner-shaped object remains the open question.
Read more at The Next Web.