What the EU AI Act’s disclosure rules mean for branded video

If you’re commissioning AI-generated or AI-assisted video for a brand, the question is no longer “do we have to disclose it?” — it’s “how, and who’s on the hook?” The EU AI Act’s transparency provisions make synthetic-media disclosure a default expectation rather than a nice-to-have. Here’s the practical version, written for the people approving the work, not the people litigating it.

What Article 50 actually asks for

Article 50 introduces transparency obligations for AI systems that generate or manipulate content. For audiovisual work aimed at the public, the principle is straightforward: if a meaningful part of what the audience sees or hears was artificially generated or substantially altered, that should be made clear to them. There are narrow exceptions, but the safe operating assumption for branded content is simple — disclose by default.

What “disclosure” looks like in practice varies by placement: an on-screen label, a caption, platform-native AI tags, or machine-readable provenance embedded in the file. Often it’s a combination.

Disclosure isn’t a tax on creativity. Done well, it’s a trust signal — and increasingly, a procurement requirement.

Who carries the obligation

The Act distinguishes between providers (those who build or supply the AI system) and deployers (those who put the content out into the world). For a typical branded video, both sides have a role:

  • The studio embeds provenance, documents the model and data choices, and flags which elements are synthetic.
  • The brand ensures the published placement — the cut that actually runs — carries the required label for that channel.

The failure mode is assuming the other party handled it. Write the responsibility split into the statement of work.

Make disclosure travel with the file

A label that lives only in a caption disappears the moment the asset is re-uploaded or clipped. That’s why provenance standards matter. Content Credentials built on the open C2PA standard sign provenance directly into the media, so the disclosure follows the file across platforms and edits. Pair that with the platform’s own AI-content labelling and you’ve covered both the human-readable and machine-readable bases.

What to do before the brief is signed

  1. Decide which elements will be synthetic, and whether they’re “substantial” enough to trigger disclosure (assume yes if unsure).
  2. Agree the disclosure method per channel, and who applies it.
  3. Require provenance/Content Credentials as a deliverable, not an afterthought.
  4. Keep the consent and rights records that sit underneath all of this — disclosure and consent are different obligations, and you need both.

None of this is exotic. It’s the difference between a campaign that ages well and one that gets pulled. At Beeyawn we build disclosure and provenance into the production pipeline by default — here’s how we handle consent and rights.

This article is general orientation, not legal advice. For a specific campaign, work with your own counsel.