YouTube adds automatic AI labels and Studio appeals for undisclosed photorealistic videos

• YouTube moved its AI disclosure for photorealistic, meaningfully altered video into the main viewing surfaces, and the YouTube announcement summary says those labels now sit below long-form players and directly on Shorts overlays.
• Official blog post says YouTube can now auto-apply a label when a creator skips disclosure but internal systems detect significant photorealistic AI use, a change that the HN thread overview immediately turned into a false-positive debate.
• According to YouTube's disclosure help page, creators still have to self-label realistic synthetic scenes, altered footage of real events, and fake speech or actions from real people.
• HN discussion highlights focused on the weak point fast: creators can fix mistaken labels in Studio, but commenters questioned whether appeals will be accurate enough for channels that depend on uploads going live cleanly.
• YouTube says in the official announcement and its help docs that the label itself does not change recommendations or monetization eligibility, but repeated non-disclosure can still trigger manual labels, content removal, or Partner Program suspension.

You can read YouTube's launch post, check the exact examples in the disclosure help doc, and see how the policy lands with creators in the main HN thread. The buried detail is that YouTube now treats three inputs as label signals: its own AI tools, C2PA content credentials, and internal detection. The weird tension is obvious in the HN replies, where the discussion summary worries at the same time about false positives and about bad actors slipping labels off.

The visible part of this update is simple. The blog post says photorealistic AI disclosures now appear directly below long-form videos and as an on-video overlay for Shorts.

That replaces the older, easier-to-miss disclosure flow with one primary label surface for realistic AI video. YouTube's viewer help page says unrealistic, animated, or lightly altered content can still live in the expanded description under “How this content was made.”

The operational change is automatic labeling. YouTube's announcement says that if creators do not disclose AI use and YouTube detects significant photorealistic AI, the platform will now apply the label itself.

In most cases, that label can be changed in YouTube Studio if the creator thinks detection got it wrong. But the same blog post and the help doc carve out non-removable cases: videos made with YouTube tools like Veo or Dream Screen, videos carrying C2PA metadata that indicate full AI generation, and some labels added after manual review.

YouTube's disclosure policy still draws the line around realistic, meaningful alteration. It gives three core cases creators must disclose:

• A real person is made to say or do something they did not say or do.
• Footage of a real event or real place is altered.
• A realistic scene is generated even though it never happened.

The same page also lists what does not need disclosure, which matters for ordinary creator workflows:

• aesthetic edits like color or lighting changes
• production assistance such as scripts, thumbnails, titles, or infographics
• captioning, sharpening, upscaling, repair, and voice cleanup
• obviously unreal or animated scenes
• cloning your own voice for voiceovers or dubs

That keeps YouTube's requirement narrower than a blanket “AI was used somewhere in production” rule. The disclosure target is realism plus the potential to mislead.

The HN reaction went straight to trust in the detector. In the thread summary, commenters questioned whether AI flagging systems are good enough to label uploads accurately, and whether appeals will move fast enough for working creators.

Three concerns surfaced repeatedly in the HN discussion highlights:

• false positives could tag legitimate videos and force creators into an appeals workflow
• creators can remove incorrect labels in Studio, which helps with mistakes
• the same removable-label path may be easy for bad actors to game

That is the whole tension of the rollout. YouTube added automation to catch undisclosed photorealistic AI, but the community response zeroed in on the same two failure modes every detection system inherits: catching too much and catching too little.

The most concrete enforcement detail sits in YouTube's help page, not the launch copy. Creators who repeatedly fail to disclose can face manual labeling, content removal, or suspension from the YouTube Partner Program.

The viewer-facing help page adds another signal path. YouTube's “How this content was made” doc says secure Content Credentials, specifically C2PA 2.1 or higher, can carry forward AI disclosures and sometimes add an “Info from” attribution naming the signing authority.

That means YouTube's new label system is not just a detector plus a checkbox. It is also a metadata pipeline, and in some cases that metadata can make the disclosure stick whether or not a creator wants to change it later.