AI disclosure rules for video in 2026: what founders need to label

Every founder I film now asks some version of the same question: "If we use AI in this, do we have to say so?"

The honest answer is: sometimes, and it's simpler than the noise suggests. The rules have tightened this year. The EU's AI labelling duty for deepfakes has applied since 2 August 2026, and YouTube and TikTok both enforce their own disclosure policies. But almost all of them turn on one idea. Get that idea right and you'll rarely get it wrong.

This is a filmmaker's practical reading of the rules, not legal advice. If AI is central to a campaign, speak to a lawyer.

Do I need to label AI in my videos?

You need to label AI when it makes something look or sound real that isn't. That is the common thread across the EU rules, YouTube and TikTok. A realistic person, place, event or voice that a viewer could mistake for genuine needs disclosure. AI that simply helps you make the video does not.

YouTube's policy covers content that "makes a real person appear to say or do something they didn't do", realistic footage of real places or events that didn't happen, and AI-generated music. TikTok's guidelines say: "You must label content that uses AI or includes significant edits to show realistic-looking people or scenes."

The test isn't "did we use AI?" It's "could someone be fooled?"

What the EU AI Act changed in August 2026

Article 50 of the EU AI Act now requires anyone who publishes a deepfake to disclose that it was artificially generated or manipulated. The European Commission defines a deepfake as AI-generated or manipulated image, audio or video that "resembles existing persons" and "would falsely appear to a person to be authentic or truthful".

A few details matter for brands:

• It wasn't delayed. The EU's "digital omnibus" pushed back the high-risk AI rules, but the transparency obligations still applied from 2 August 2026. Only a narrow technical marking duty for AI providers got extra time, to December 2026.
• Creative work gets a lighter touch. For evidently artistic, creative, satirical or fictional work, the disclosure can be made in a way that doesn't spoil the work, according to Paul, Weiss's summary of the Commission guidelines. Wholly fantastical content that nobody would take as genuine isn't a deepfake at all.
• It can reach UK businesses. The Act applies to deployers outside the EU where the output is used in the EU. Being viewable from Europe isn't enough on its own, but if you market to EU customers, plan as if it applies.
• Fines are real. Up to €15 million or 3% of worldwide annual turnover, with proportionality for smaller companies.
• Nothing is retroactive. Deepfakes made before 2 August 2026 don't need labelling after the fact, though the Commission encourages it.

What YouTube, TikTok and Instagram require

The platforms got there first, and their rules are what most founders will actually run into day to day.

Two things reassure most clients. YouTube says disclosure alone doesn't limit a video's audience or monetisation eligibility. And TikTok lets you use its built-in AIGC label or your own clear caption, sticker or watermark.

What about the UK?

There's no blanket legal requirement in the UK to disclose AI in ads. The ASA says the CAP and BCAP Codes have no AI-specific rules. The usual rule applies: an ad must not mislead.

The ASA's test is a good one for anyone to borrow: "Is the audience likely to be misled if the use of AI is not disclosed?" It also notes that routine editing, the kind post-production has always done, doesn't need flagging just because AI did it.

Its most important line is this: "disclosure alone is very unlikely to mitigate the harm caused by a fundamentally misleading message." An AI-perfected skin result with a small "AI" tag is still a misleading skincare ad.

A five-question check before you post

This is the check we run on our own edits. It takes a minute.

1. Is any person, place, event or voice in this generated or meaningfully altered by AI? If not, you're done. Grading, captions, noise reduction and AI-assisted editing don't count.
2. Would a reasonable viewer think it's real? A stylised animated explainer, no. A realistic shot of your founder on a stage they never stood on, yes.
3. Does it involve someone other than you? A cloned client voice, a generated customer or an altered colleague needs a label and, frankly, written consent.
4. Does the content still tell the truth? If the AI element makes a claim about results, products or people that isn't true, fix the content. A label won't save it.
5. Is it labelled where people will see it? Use the platform's own toggle where there is one, and add a line in the caption for EU-facing content. Keep a note of what was generated in the project file, so you can answer questions later.

How we handle it on our shoots

Our rule is simple. Real people, real places, real words. We use AI where it saves time without changing the truth: transcription, rough cuts, clean-up, upscaling, the odd extended background. When something synthetic stands in for something real, we say so on screen or in the caption, and we get permission from anyone whose face or voice is involved.

It isn't only about compliance. Your audience follows you because they believe it's you. Every video that later turns out to be quietly fake spends some of that belief. A one-line label costs nothing by comparison.

If in doubt, label it. Nobody unfollows a founder for being honest.