YouTube said it was rolling out internal detection signals that could automatically label videos containing substantial photorealistic material made or altered with artificial intelligence. The change extended a disclosure system introduced in 2024, when the platform began labeling content after creators reported their own use of AI tools.
The automatic process was scheduled to begin during May 2026. If a creator did not provide an AI disclosure but YouTube’s systems detected significant realistic generation or manipulation, the service could add the label itself. The supplied announcement did not specify detection thresholds, error rates or the full range of signals used.
Creators still carried a requirement to disclose realistic AI use manually. They could change the status in YouTube Studio if they believed an automatic identification was wrong. YouTube said disclosures would remain permanent in a limited set of cases, although the supplied excerpt did not include the categories covered by that exception.
The company also changed where viewers would see notices. A single disclosure format for photorealistic and meaningfully generated or altered material was moved into a more prominent position on the primary viewing interface. YouTube described that placement as a way to provide context immediately.
Different treatment applied to less realistic material. Disclosures for animation, obviously fictional images or slight alterations remained available in the expanded video description rather than receiving the prominent on-screen placement. The distinction focused attention on media viewers might plausibly interpret as a record of real people or events.
YouTube said community feedback had consistently favored greater transparency around generative content. Its two-part update combined a clearer visual location with platform-initiated labeling, reducing exclusive reliance on a creator’s decision to disclose. At the same time, the correction mechanism was intended to preserve some creator control as detection technology evolved.
A label by itself would not change whether a video could earn revenue, the company said. It also would not alter the recommendation system’s treatment of the video. The notice was therefore presented as information for viewers, not an automatic penalty or distribution restriction.
The announcement established the policy and rollout timing, but it did not show how often automatic labels had already been applied or how accurately the system distinguished generated footage from conventional production. Those operational questions remained unanswered in the supplied evidence.
As of the date, the practical shift was clear: realistic AI content on YouTube could receive a disclosure even when its uploader supplied none. Viewers would see that notice more readily, and creators could dispute most automated decisions through Studio without the label alone affecting reach or income eligibility.
