Article 50 of the EU AI Act has applied since August 2, 2026 – and with it the transparency obligation for AI content. Anyone publishing AI-generated or AI-manipulated images, video or audio must disclose their origin; in addition, the content has to carry a machine-readable marking identifying it as AI-generated. For brands and agencies working with AI imagery, this has been binding law since that date – no longer a future topic.
Anyone hoping for a last-minute delay is out of luck: the Digital Omnibus (Regulation (EU) 2026/1744) did not postpone the transparency obligations of Article 50. They apply – and Germany has already set up enforcement.
What has to be labeled
The obligation works on two levels. First, disclosure to the audience: anyone advertising with deepfakes – AI content that realistically depicts real people or looks like authentic footage – must make the AI origin clearly recognizable. Second, the technical level: AI-generated content must be marked in a machine-readable way so that systems can automatically identify it as such.
The European Commission's guidelines of July 20, 2026 clarified two points that are decisive for advertising. Labeling is also required for fictional people who could realistically exist – which is exactly the kind of AI model many campaigns work with today. And the exception for artistic works practically never applies to advertising.
What the labeling should look like in practice is described by the Wettbewerbszentrale, Germany's competition watchdog, in its guide (updated July 29, 2026): AI models with human-like personas must be labeled, the label must be in German for a German audience, and vague wording such as “AI-supported” is not enough.
Who enforces the rules – and what violations cost
Germany organized enforcement at the end of July 2026: the KI-MIG – the German act implementing enforcement of the AI Act – entered into force on July 29, 2026 and makes the Federal Network Agency (Bundesnetzagentur) the central complaints office. Anyone can report unlabeled deepfakes there.
In parallel, the Wettbewerbszentrale has been running its own AI complaint form since July 28, 2026. And there is a third lever: legal commentary (including TWW and LTO) classifies Article 50(4) as a market conduct rule within the meaning of Section 3a of the German Unfair Competition Act (UWG) – missing labels can therefore also be challenged by competitors with cease-and-desist letters. On top of that, the AI Act's fining framework under Article 99(4)(g) reaches up to 15 million euros or 3 percent of global turnover.
The courts have set the tone
German courts drew clear lines even before the deadline. On March 4, 2025, the Higher Regional Court of Frankfurt am Main ruled (case 16 W 10/25) that Meta must proactively block deepfake ads with equivalent content – removing only the one reported ad is not enough. And on August 20, 2025, the Regional Court Berlin II awarded a claimant 4,000 euros (case 2 O 202/24) because an AI voice clone violated his personality rights.
What the platforms are already doing
The platform side has organized itself as well. The Code of Practice on Transparency, which fleshes out how Article 50 is implemented, has been final since June 10, 2026; by July 31, 2026 around 190 companies had signed it – including OpenAI, Google, Meta, Microsoft, Lufthansa and Getty.
Labeling has arrived in the ad systems too: Google Ads rolled out settings for AI labels in July 2026 (for the EU, India and New York), and Meta has been automatically labeling ads since June 1, 2026. You should not rely on that alone, however: the Wettbewerbszentrale expects a clear, German-language label on the ad itself – an automatic platform label does not replace your own diligence.
Compliance as an advantage: licensed and labeled
How do audiences react to openly labeled AI advertising? Better than many in the industry believe – just differently than they assume. According to a survey by IAB and Sonata from January 2026, 82 percent of advertising decision-makers assume consumers view AI ads positively; in reality only 45 percent do. At the same time, 73 percent of consumers say that open disclosure does not reduce their intent to buy. Transparency, in other words, costs hardly any effectiveness – while a missing label has been a tangible legal risk since August 2, 2026.
This is exactly where FaceLedger comes in. On the marketplace, real people license their likeness explicitly for AI advertising – per campaign, with documented consent that can be revoked at any time. The labeling required under the AI Act is built into every license, and the terms are transparent: every license starts at 900 euros net, with 65 percent going to the person depicted. For you, that means cleared rights, provable consent and AI Act-compliant labeling in a single step – turning an obligation into a quality signal you can show clients and consumers.
Since August 2, 2026, labeling AI advertising has been mandatory – with three enforcement routes: the Federal Network Agency, the Wettbewerbszentrale and cease-and-desist letters under the UWG. Licensed likenesses with built-in labeling turn this obligation into a trust advantage.
Every AI-generated face is a legal risk. Unless it is licensed.
Request agency access