YouTube adds automatic AI labels below the player and in Shorts
YouTube says photorealistic or meaningfully AI-altered videos will get more prominent disclosures and may be auto-labeled with internal signals. Creators can contest mistakes, but Veo-made outputs or files with C2PA-style metadata may keep permanent labels.
TL;DR
- YouTube is moving AI disclosures for photorealistic or meaningfully altered video into places viewers actually see, according to YouTube's policy summary, with long-form labels below the player and Shorts labels as an on-video overlay.
- Starting in May 2026, YouTube says it will also auto-apply labels when its systems detect significant photorealistic AI use that a creator did not disclose, as described in the YouTube announcement and discussed in the main HN thread.
- Creators can usually change a mistaken label inside YouTube Studio, but YouTube's announcement says labels stay permanent for videos made with Veo or Dream Screen, and for files carrying qualifying C2PA metadata.
- The HN discussion roundup focused on the messy part: false positives, mixed workflows, and whether documentary-style videos that combine AI voice, generated shots, and archive footage will get swept in.
YouTube's own post says viewers will now see the label directly below the player or on the Short itself. The companion help docs spell out the edge cases, including which edits do not need disclosure and how Content Credentials can carry AI provenance into YouTube. Meanwhile the HN thread immediately jumped to the creator problem: detection is useful right up until it hits the wrong channel.
Label placement
YouTube Updates AI Labeling for Improved Transparency and Automated Detection
YouTube has announced updates to its AI labeling system to improve transparency and ease of use for viewers and creators. Beginning in May 2026, disclosures for photorealistic or meaningfully AI-altered content will appear in more prominent locations: directly below the video player for long-form videos and as an on-video overlay for Shorts. Additionally, YouTube is introducing internal signals to automatically detect and label significant photorealistic AI content if creators fail to disclose it. While creators retain the ability to update labels if they believe their content was misidentified, labels remain permanent for content created using YouTube's own AI tools (e.g., Veo, Dream Screen) or containing specific generative AI metadata (e.g., C2PA). Content that is unrealistic, animated, or only slightly altered will continue to have disclosures located in the expanded video description.
The core product change is placement. YouTube says the single disclosure format for photorealistic and meaningfully AI-altered video now sits directly below the player on long-form uploads, while Shorts get the disclosure as an overlay on the video itself, per the official blog post.
The old, easy-to-miss description treatment still exists for a different class of material. YouTube's help page says non-photorealistic, animated, or only slightly altered content can keep its disclosure inside the expanded description instead of the main player surface, according to the disclosure policy.
Auto-detection
YouTube to automatically label AI-generated videos
If you make video content, this matters because YouTube is changing how AI disclosure is surfaced and is starting to auto-label some photorealistic AI content. The discussion focuses on creator-facing consequences: false positives, appeal workflows, and whether AI-heavy channels can be detected even when the output is mixed with human editing or archival footage.
YouTube says it is rolling out "internal signals" to identify significant photorealistic AI use when creators leave the upload disclosure blank, according to the official announcement. If the system flags a video, the platform can apply the label automatically.
The company also says the label alone does not change recommendation ranking or monetization eligibility, per the same post. That is the clean product story. The HN discussion roundup surfaces the harder one: commenters immediately worried about false positives and asked how reliably any AI detector can separate synthetic shots from heavily edited real footage.
Disclosure rules
Discussion around YouTube to automatically label AI-generated videos
Thread discussion highlights: - numpad0 on Detection accuracy: Asks about the overall state of AI-based AI flagging tools and notes they can have very high false-positive rates. - CM30 on Creator risk vs. abuse: Says automated flags can wrongly hit innocent creators, but also argues the new labeling is needed because AI-generated videos are easy to hide from viewers. - leoc on What content gets labeled: Questions whether YouTube's sweep will catch AI-made math, physics, CS, and documentary-style videos that mix AI scripts, voices, and generated visuals with archive footage.
YouTube's help docs break the policy into a short list of realistic cases that require disclosure:
- A real person appears to say or do something they never did, per YouTube Help
- Footage of a real place or event is meaningfully altered, per YouTube Help
- A realistic scene is generated for something that never happened, per YouTube Help
The same page also draws a line around what does not need disclosure. Minor beauty filters, color correction, background blur, caption generation, script or thumbnail assistance, upscaling, repair, and even cloning your own voice for voiceovers or dubs are listed as non-disclosable examples in YouTube's guidance.
That gap is where the creator anxiety lives. In the HN roundup, one commenter asked whether math, physics, computer science, or documentary-style channels that mix AI scripts, voices, generated visuals, and archival footage would get caught by the same sweep.
Permanent labels
YouTube Updates AI Labeling for Improved Transparency and Automated Detection
YouTube has announced updates to its AI labeling system to improve transparency and ease of use for viewers and creators. Beginning in May 2026, disclosures for photorealistic or meaningfully AI-altered content will appear in more prominent locations: directly below the video player for long-form videos and as an on-video overlay for Shorts. Additionally, YouTube is introducing internal signals to automatically detect and label significant photorealistic AI content if creators fail to disclose it. While creators retain the ability to update labels if they believe their content was misidentified, labels remain permanent for content created using YouTube's own AI tools (e.g., Veo, Dream Screen) or containing specific generative AI metadata (e.g., C2PA). Content that is unrealistic, animated, or only slightly altered will continue to have disclosures located in the expanded video description.
The sticky part is provenance. YouTube says creators can usually correct a mistaken AI disclosure in Studio, but not when the video was made with YouTube's own tools such as Veo or Dream Screen, or when the file includes C2PA metadata showing the entire video was fully generative AI, according to the blog post and the disclosure help page.
A second help document adds that YouTube carries forward secure Content Credentials from C2PA 2.1 or higher, can show an "Info from" attribution for the signing authority, and may place that provenance either on the player or in the detailed description, per How this content was made. The underlying C2PA explainer describes those credentials as cryptographically bound provenance records, which is why some labels are now attached to the file history rather than just to a creator's checkbox.