Rules and practice

AI ads are allowed. Since August, some of them have to say so.

Article 50 of the EU AI Act started applying on 2 August 2026, and the platforms have their own rules on top. Here is what has to be labelled, what does not, and how we handle it on the campaigns we run.

Magnet Minds 20 September 2026 7 min read

AI-generated video went from a curiosity to a normal line in a media plan in about a year. The part that changed this summer is not whether you may use it - you may - but whether the viewer has to be told. Since 2 August 2026 that question has a legal answer in the EU, and the platforms have their own answers on top of it.

This is the practical version: which duty falls on the tool, which falls on you, and what we actually do on the campaigns we run.

What changed on 2 August 2026

Article 50 of the AI Act - the transparency article - started applying on 2 August 2026. The European Commission describes it in two halves, and the split matters, because only one half is your problem.

2 Aug 2026

Article 50 started applying

Two duties

one on the tool, one on the advertiser

15m EUR / 3%

the ceiling for breaches of Art. 50 (Art. 99(4))

Art. 50(2) - the tool

The provider marks the file

Whoever supplies the AI system that generates synthetic image, audio, video or text marks the output in a machine-readable format, detectable as artificially generated. That is the generator's job, not yours. Assistive editing that does not substantially alter the input is exempt. Systems already on the market before 2 August got a four-month transition for the marking.

Art. 50(4) - you

The deployer discloses the deepfake

Whoever uses an AI system to create a deepfake discloses that the content is artificially generated or manipulated. In an ad, the deployer is the brand or the agency running it. Art. 50(5) adds the how: clearly and distinguishably, at the latest at first exposure. A line in a landing page footer does not satisfy that; a visible marker on the creative does.

One thing worth knowing, because it caused a lot of noise: the Digital Omnibus on AI, Regulation (EU) 2026/1744, in force since 27 July 2026, pushed the high-risk deadlines back to December 2027 and August 2028. It did not move Article 50. The transparency date stayed where it was.

What counts as a deepfake, and what does not

"Deepfake" in the AI Act is broader than the word suggests in everyday use. The definition covers AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events, and that would falsely appear to someone to be authentic.

Read that against a normal ad and the line is workable.

Treat as disclosable

A generated person who reads as a real customer
A generated shot of a shop or street that looks real
A scene that looks filmed but never was
Real footage altered so it shows something that did not happen

Not a deepfake

An obviously unreal world - a product flying through an impossible landscape
Colour grading, cropping, cleanup, upscaling
AI help with the script or the edit, on real footage
Artistic, creative or satirical work - lighter regime, disclosure that does not spoil the work

The test is not "did AI touch this" but "could a viewer take it for a record of something real".

The Commission published final guidelines on these obligations alongside a voluntary Code of Practice on Transparency of AI-generated Content, which providers and deployers can use to show they comply. Neither replaces reading the article itself.

Not legal advice

This is a summary of the text of the AI Act and of published platform policies, not advice on your specific campaign. In Bulgaria the national coordinator for the AI Act is the Communications Regulation Commission (CRC).

What Meta, TikTok and YouTube require

Platform rules are separate from the law and enforced faster. A rejected ad arrives the same day; a regulator does not.

TikTok

Ads Manager treats AI-generated, synthetic or manipulated media as a mandatory disclaimer: content that is fully AI-generated, or real footage significantly modified by AI. The disclaimer is set in Ads Manager when the ad is built, rather than left to be discovered later.

YouTube

Creators must disclose realistic altered or synthetic content - a person made to appear to say something they did not, a real place shown in an event that never happened. A label shows in the player. Cosmetic edits, beauty filters, colour adjustments, upscaling and AI help with scripting do not require disclosure, and disclosing does not limit the audience or monetisation.

Three separate rulebooks. The AI Act sets the floor; the platforms enforce their own on top of it.

AI video ads: what labelled work looks like

These are ours, all generated. Nothing about a disclosure changes how they look: the label sits in the ad's information, not across the picture.

Car giveawaygenerated cast, locked palette

Claw machineone idea, six shots

Pink prioritysame face across the set

Ours, from the mAInds portfolio. See the rest of the reel.

Want AI video that ships without surprises?

We write, generate, edit and deliver the videos - and hand them over labelled, with the provenance kept, so the ad account never trips over it.

See the AI video packages

How we handle it on live campaigns

None of this makes AI video harder to sell. It makes the paperwork part of the job, which is where an agency earns its fee. Our working order:

Before the shoot list

Decide per concept whether it depicts something that could be taken as real. That decision drives everything downstream, so it happens before generation, not after.

During production

Keep the provenance metadata the generators write. Re-encoding a clip carelessly strips it, and that is the signal platforms read.

On delivery

The disclosable creatives go out with the platform's own AI declaration set, plus a visible marker where the concept calls for it.

In the ad account

The declaration is set in the ad, not left for automated detection to guess at. Detection is a safety net, not a strategy.

On record

A short log of which clips were generated, with which tool, for which campaign. If anyone ever asks, the answer takes a minute.

If you are planning AI ads but have no creative yet, start with AI video, and for running them with the ads page. The longer version of how the videos get made is in how to make an AI video. Our own statement on where we use AI is on the AI transparency page.

Questions

If the creative is a deepfake in the sense of the AI Act - AI-generated or manipulated content resembling real persons, objects, places or events that would falsely appear authentic - then yes, the deployer has to disclose it, clearly and at the latest at first exposure. An obviously unreal, fantastical scene is a different case, and assistive edits like colour grading are not covered.

No. The AI Act duty on advertisers is about deepfakes, not about every use of AI. Separately, the platforms label more broadly on their own: Meta applies an AI info label to ads made with its AI tools and to third-party AI it detects, and TikTok requires a disclaimer for AI-generated or significantly AI-modified media in ads.

Under Article 99(4) of the AI Act, breaches of the transparency obligations in Article 50 carry administrative fines of up to 15 million euro or 3% of total worldwide annual turnover, whichever is higher. For SMEs and start-ups the cap is the lower of the two. In Bulgaria the national coordinator is the Communications Regulation Commission.

Not as far as the platforms state it. YouTube says explicitly that disclosing synthetic content does not limit a video's audience or its eligibility to earn money, What does hurt performance is a rejected ad.

Want AI video that ships clean?

Look at what the packages include - concept, script, generation, edit, sound and two rounds of revisions - and what each one costs.

See the AI video packages

Sources