Most agency owners I speak to in Europe have heard of the AI Act and have not read it. That is not an insult — it is a 400-page regulation written for lawyers, and they have client work to ship.
So here is the part that matters for video, and the part that does not.
What matters: the transparency obligations for synthetic content took effect on 2 August 2026. If an ad depicts a person or a scene that does not exist, and a viewer could reasonably believe it does, that ad is expected to carry a disclosure. That is a legal duty sitting on the brand, and a professional one sitting on the agency that supplied the creative.
What does not matter to you: the risk-management architecture, the conformity assessments, the high-risk classification lists. None of it touches a 20-second skincare reel. Agencies talk themselves out of the useful part by worrying about the irrelevant part.
What actually counts as synthetic
In practical terms, for advertising work, there are four cases:
| Case | What it looks like | Disclosure position |
|---|---|---|
| Synthetic person | An AI-generated face or voice presenting the product | Expected to be disclosed |
| Real product, invented scene | A real SKU composited into a lifestyle setting it was never filmed in | Lower risk, depends on presentation |
| Stylised or clearly graphic | Motion graphics, illustration, obvious animation | Generally outside the intent of the rule |
| Real footage, AI edits | Real shoot, AI captions, colour, cleanup | Not synthetic content |
The dividing line is whether a reasonable viewer could be misled about what exists. A cartoon explainer violates nothing. A woman who never existed, telling you a serum works, is the case the rule was written for.
The platform label is not the legal disclosure
This is where agencies get caught, and it is worth being precise.
Meta, TikTok and YouTube each run their own synthetic-content labelling and each fires on a different trigger, with different detection and different wording. Your asset may get a platform label, or it may not, depending on how it was produced and what was declared at upload.
That label is a platform policy artefact. It does not discharge the legal duty, and the absence of one is not evidence of compliance. They are two separate obligations that happen to point the same way.
So what do you actually do on Monday
Not a policy document. A disclosure sheet per delivered asset — five fields, attached to the batch, kept with the client file:
1. Was a synthetic person used? Yes or no. Not "AI-assisted" — a specific answer.
2. Was the product real, and was it filmed or composited? The distinction a complaint would turn on.
3. Which model or pipeline produced the visuals? Named, so the chain is traceable.
4. Does the platform require a label on upload? With the platform named.
5. Was a human reviewer satisfied the result is not misleading? A name and a date.
Five fields take about four minutes per asset. The point is not the paperwork. The point is that when a client's legal team asks in month nine what was generated and how, somebody can answer in one email instead of reconstructing it from Slack.
The three mistakes I see
Treating the caption as the disclosure. A disclaimer in small type at second fourteen is not disclosure; it is set dressing. Disclosure has to be attached to the asset and legible for its duration, or it is a defence that will not survive contact.
Assuming the platform will handle it. Platforms label what they detect or what you declare. They do not audit your legal position, and a mislabelled asset does not become compliant because a distributor applied a tag.
Keeping nothing. The most common failure is not a bad disclosure, it is no record at all. An agency that can show a dated five-field sheet for every synthetic asset has a defensible position almost regardless of the interpretation. An agency that cannot has only its own recollection.
What this changes about buying video
If you resell production, this is now a supplier-selection criterion, and a good one:
- Ask any production partner what they hand over alongside the file. If the answer is nothing, that is the answer.
- Ask whether they will state, per asset, whether a synthetic person appears. A producer who cannot answer that from their pipeline does not have a pipeline, they have a prompt.
- Put the disclosure duty in the contract, not the email thread.
Where NNI sits
Every asset we deliver ships with a disclosure sheet on those five fields — model used, synthetic-person status, product treatment, the platform label it needs, and the reviewer's name. It sits beside the file, it goes in your client folder, and it is written in language a brand's legal team can use without a translation pass.
That is not a premium tier. It is what a synthetic-content pipeline should have shipped with from the start.
If you are selling into the EU and want to see what the sheet looks like before you order anything, ask and I will send one.
---
Jayesh runs NNI Design Studio, white-label video production for agencies. Rate card at nnidesignstudio.com/pricing; European delivery notes on the Europe Block.
Drop your take 0
// no hot takes yet. start the thread 👇