Tags:
Innovation LabNotes from the Innovation Lab is a monthly roundup from the Freedman Innovation Lab, on the AI, technology and process shifts that matter in global marketing.
The final countdown on the EU AI Act
The deadline is looming for the EU AI Act's transparency obligations to take full effect, and this regulation lands specifically on advertising. It applies to every marketing campaign regardless of the size of your budget, and the penalties for getting it wrong are steep: up to 15 million euros or 3% of global turnover, whichever is higher.
From the 2nd of August 2026, any AI-generated marketing asset that runs in the EU has to satisfy two obligations:
1. A machine-readable marker in the file’s metadata
The genAI systems that generate synthetic image, video, audio or text must mark their outputs in a machine-readable way, so the file itself signals it was made or altered by AI. This is not visible on the image, but hidden information built into the file.
Every digital photo already carries hidden data you never see on the image, such as the camera, date and settings. The AI version is similar: a tamper-proof, signed note built into the file, recording that AI was involved, which system produced it, when, and what changed afterwards. The major technology and media companies have agreed on one shared open standard, so the same markings can be read by different tools and platforms.
2. A noticeable label wherever the content appears
The label pairs two things: the standardised EU "AI" icon and a short line of text such as "Generated with AI" or "Manipulated with AI". There is no single mandated sentence - the Act is written in principle, and the detail sits in the accompanying Code of Practice. Disclosure is mandatory, but the EU icon set is not, and the icon alone does not prove compliance. It is simply the safest ready-made option. The sensible default is to use the official icon unmodified, add a short line, and translate only the words for each market. The icon adapts on its own, so "AI" becomes "IA" in France and Italy and "KI" in Germany.
The label can sit on the asset or around it, whether using a platform's AI tag or adding a line in the caption. On size and placement there is more freedom than people assume: the Code sets no fixed pixel size and names no particular corner of the frame. The firm rule comes from the Act: disclosure must be clear and distinguishable at the first interaction or exposure. In practice that means easy to see the moment the content appears, and legible. Not buried on the second frame of a carousel or hidden at the bottom of a caption nobody expands.
But the rules for video and audio are more specific. On a short video, which covers most ads, the label should stay on screen throughout rather than flash once, because people scroll in halfway. On longer video, open with the disclosure, keep a small persistent icon, and repeat it. Audio has a threshold: under 30 seconds, a spoken line at the start; over 30 seconds, repeated partway through.
How will the EU AI Act show up in a real campaign?
-
A TV and online spot where a real actor is re-voiced into German by a cloned voice: a short label at the head of the film, and the markings embedded in the master file itself, so the record that it was AI-altered travels with the file rather than living in someone's notes.
-
A social cutdown where the talent's lips are reanimated to match the dubbed track: because it is a short video, the on-screen label stays visible throughout, since viewers arrive mid-clip, reinforced by the platform's AI tag and a line in the caption, with the file's markings built to survive a re-upload that would otherwise strip them.
-
A printed poster is the opposite case. A digital out-of-home screen behaves like video and carries markings in its file, but a physical print carries none; there is no file to inspect. And because a realistic AI-generated image counts as a 'deepfake' under the Act, even a made-up but natural-looking scene or person, the visible label is required anyway. So on print you lose the invisible layer entirely, the label does all the work.
The real challenge for global marketers
Here is the sting in the tail. A single EU-wide law sounds like it should make life simpler, one rulebook instead of twenty-seven. But for international campaigns it adds yet another layer to an already complicated set-up, requiring new considerations by location, format and channel in addition to the existing complexity of language, cultural adaptation, media specs, market approvals and other existing regulations.
The governance gap in global campaigns makes this harder. On McKinsey's 2025 figures, 71% of organisations now use generative AI, yet fewer than a third follow most established practices for governing it. Adoption has run well ahead of the discipline for handling it. The organisations that take the 2nd of August deadline in their stride will be the ones that have already made disclosure part of normal production, rather than piecing it together asset by asset at the last minute.
What is the best way to apply the EU AI Act across markets?
Thirty markets inventing thirty slightly different answers, on where the label sits, how big it is, what colour and what it says, does not produce thirty compliant assets. It produces thirty inconsistent ones, and that inconsistency quietly risks exposure to the penalties of the Act.
The fix is to decide once, at the mastering stage. Build a defined space for the AI label into the master, settle the wording, placement, size and colour there, and let it cascade identically through every European version of a 360 campaign. The icon stays constant and only the translated text changes. One approach, decided once, carries the whole campaign. The Act helps here: it is clear on what it requires but deliberately open on how you execute it, and that latitude is exactly why a single agreed approach, fixed at the master, is worth having.
That decision extends beyond the EU. If the same master runs in a market outside Europe with no equivalent regulation, there is no legal requirement to carry the label at all, so it becomes a brand choice rather than compliance. Some brands will want transparency everywhere for consistency, others will not. Either way the decision should be deliberate and recorded.
From now on, the invisible markings and the visible label will be part of the asset, like a colour profile or a safe area. They travel with the file itself. But while the technology can mark the file, it cannot decide what counts as meaningfully AI-generated, what a market expects, or whether thirty versions tell the same story about how they were made. That responsibility remains with the global marketing team.
There is still time to build the right approach for your brand, but the window is narrowing. At Freedman, we have the expertise to guide you through the new rules and keep your global campaigns compliant without losing what makes them land in each market. If you need help implementing the EU AI Act across your European campaigns, get in touch today.
Sources:
Article 50: Transparency Obligations, EU Artificial Intelligence Act
