mırakı DEENFRIT
Miraki Mind

AI labelling: what 2 December means for brands

Since August, the EU requires AI content to be labelled. On 2 December a grace period ends that is often misread. What was pushed back, and what brands should check now.

Miraki MindAI-assisted editorial · 26 October 2026 · 8 min read

Since 2 August 2026, Article 50 of the EU AI Act applies: the transparency obligations for AI-generated content. The AI Omnibus, Regulation (EU) 2026/1744, in force since 27 July 2026, added a grace period: providers of generative systems that were already on the market before 2 August only have to mark their outputs in a machine-readable way by 2 December 2026. That's a little over five weeks from now.

We make films with generative AI, so the labelling question comes up on every project. We label our own work: every case page ends with a note on what was made with generative AI and what was made by people. Spec ads carry the white "Spec Ad" pill and a disclaimer that the brand neither commissioned nor approved the film. For clients, we add the label on request. The responsibility, though, arises with publication, and it lies with whoever publishes the film. If it also runs outside Switzerland, in the EU, the labelling obligation of the AI Act applies there (more on that below).

Our impression: so far, little gets labelled, and the obligation is often not taken seriously. That's why we want to draw attention to it.

In our process, "Labelling under the EU AI Act" is part of the last step, the master. For exactly this question, we built the AI Act Navigator: a free check in the Lab (in German) that gives a first assessment of whether and how a film needs a label, with the deadlines after the Omnibus and the official labels to download.

What is actually new

2 December is not a new obligation and not a delay of labelling as a whole. According to the version of Article 111(4) published by the European Commission, the deadline only covers Article 50(2), only providers, and only systems placed on the market before 2 August 2026. Everything else has applied since August.

The regulation separates two layers, and people mix them up a lot:

  • The invisible mark (Art. 50(2)): Providers of systems that generate images, video, audio or text make sure the outputs are marked in a machine-readable format and detectable as AI-generated. A purely assistive function for standard editing is exempt. This is the only place the grace period applies.
  • The visible notice (Art. 50(4)): Anyone publishing deepfakes has to disclose that they are artificially generated or manipulated. For evidently artistic, fictional or satirical work, a notice that doesn't hamper the display is enough. This duty has applied since 2 August, no delay.
Marked in a machine-readable way and detectable as AI-generated: that is what Art. 50(2) asks of providers

On top of that, the Commission offers two aids. The Code of Practice on Transparency of AI-generated Content has been final since 10 June 2026: section 1 covers marking and detection for providers, section 2 the labelling of deepfakes and certain texts for deployers. It is voluntary. By the end of July, about 190 organisations had signed it, according to the Commission. And there are EU icons: a basic icon, one for "fully AI-generated" and one for "partially AI-modified", each in black and white, solid or semi-transparent. Free to use, no attribution needed.

The use of these icons does not establish legal compliance by itself.

European Commission, page on the EU icons

What it means for a production

The mark under paragraph 2 sits with the system that generates the content, not with the studio using it. Still, a production faces a practical question: do the tools a film is made with mark their outputs? And does a mark survive when the material goes through edit, grade and encoding? That's something we want to know exactly for our own tools.

The visible notice, on the other hand, is craft. The Commission recommends showing the icon clearly at the latest at first exposure, with nothing layered on top, and in a way that stays visible when content is shared or downloaded. Creative works get more room as to where the notice goes. For us, that means the label comes with the master, for every version and every channel, not as an afterthought.

Example: a frame from 21 ZUAN with the EU icon for fully AI-generated content. The woman looks real but is invented. A spec ad, not commissioned or approved by Seagull 1963

What it means for a brand

First things first: this is not legal advice, just a list of what's worth a look before 2 December. If in doubt, a lawyer should clear it.

  • Know your role: If you use AI tools professionally and publish content with them, you are a deployer. If you develop a system that generates content and place it on the market under your own name, you are a provider. Only providers with a system that was on the market before August need to care about 2 December.
  • Spot deepfakes: Does AI material resemble real people, objects, places or events, and could it falsely pass as real? Then the regulation treats it as a deepfake, and it has needed a notice since August.
  • Ask around: Which tools do your agency and studio use, and do those tools mark their outputs?
  • Pick icon and placement: per channel, per format, before the campaign goes out.

And Switzerland? The AI Act also applies to providers and deployers outside the EU as soon as the output of an AI system is used in the EU (Art. 2(1)(c)). A campaign that also runs in Germany, France or Italy is covered. At home, the Federal Council decided on 12 February 2025 to ratify the Council of Europe's AI convention; the Federal Department of Justice and Police is to draft a consultation bill by the end of 2026, covering transparency and data protection among other things.

Our take and the limits

2 December is mainly a date for the tool providers. For brands, the visible notice is the bigger question, and it already applies. If you have been open about what was made with AI all along, there's little to rebuild. And a notice that's planned in from the start takes nothing away from the film.

A lot is still open. How robust marks are in practice remains to be seen. Whether the code of practice counts as adequate is only decided once the Commission and the AI Board have assessed it. And in Switzerland there is no bill on the table yet. Our navigator gives a first assessment, but it doesn't replace legal advice.

Sources

Have an idea that was too big?

Tell us about it in a few lines. We reply within one working day.hello@miraki.ch+41 31 348 90 90
Start a project