TL;DR: From August 2, 2026, the EU AI Act says AI-generated and AI-altered content in ads has to be labeled when it could pass for real. Part of that happens automatically inside your software. The part you own is a plain, visible "made with AI" note on anything that looks real but is not, when the ad reaches people in the EU (EU AI Act, Article 50, 2026). It does not ban AI in ads. Skipping the label can cost up to 15 million euros or 3% of global turnover. I am a photographer, not a lawyer, so treat this as a researched plain-English map, not legal advice. When a case is fuzzy, the safe move is simple: label it.
On August 2, 2026, the EU's new rules on AI-made content start to apply. If your advertising uses AI to create or alter images or video, and those ads reach people in the European Union, you are now expected to be upfront about it. This is a real regulation with real fines, not a voluntary badge.
I run a hybrid production studio out of New York and London. We shoot real food, real drinks, real spaces, and real people, then use AI to stretch that work into more formats and more campaigns. So these rules land straight on my desk, the same way New York's AI ad law did earlier this summer. Here is the plain version: what the law asks for, who it applies to, what actually needs a label, and what to do about it. Written for marketers, not lawyers.
Two ground rules before we start. First, I am not a lawyer and this is not legal advice. It is researched, sourced, and current, but if a campaign has real money or real risk on it, run it past counsel. Second, here is the rule of thumb that keeps you out of trouble: when you are not sure whether something needs a label, label it. Disclosure is cheap. Getting caught hiding AI is not.
One thing to clear up right away, because the reporting has been a mess. A 2026 package delayed parts of the AI Act. It did not delay this part.
When Does the EU AI Act Take Effect?
The law arrives in stages, not all at once. It became law on August 1, 2024, and the pieces switch on over a few years (European Commission, 2026). The banned uses of AI came first, in February 2025. The rules for the big AI models came in August 2025. The part that matters for advertising, the duty to label AI content, starts August 2, 2026, across all 27 member states.
Now the part that has caused confusion. In late 2025 the EU proposed a package to simplify the AI Act, and it moved fast, with final sign-off from the Parliament and Council in June 2026 (European Parliament, 2026). It pushed back the heaviest rules, the ones for high-risk AI systems, by more than a year. It left the labeling rules alone (Gibson Dunn, 2026). So if you have read that "the EU delayed the AI Act," read it again. The AI content labeling deadline is on time: August 2, 2026.
What Does the EU AI Act Actually Require?
Two jobs, and only one of them is really yours.
The first job belongs to the software. When an AI tool makes or changes an image or a video, the tool is supposed to tag that file behind the scenes so it can be recognized later as AI. Think of it as invisible ink baked into the file. This is already real: Google's SynthID embeds that kind of invisible marker into AI-made images, video, and audio, and other AI companies have started adopting the same system, so the tag increasingly rides along with the file on its own (Google DeepMind, 2026). You do not create that tag, the tool does, and over the next year expect it to become a standard feature of the AI tools you already use. Your part here is small: choose tools that do it, and do not strip that hidden tag out when you export.
The second job is yours, and it is the one to focus on. When you publish AI content that could pass for real, you have to tell the audience it was made or altered with AI, in a way they can actually notice, the moment they first see it (EU AI Act, Article 50, 2026). That is the whole ask. A clear, visible note. The law leans hard on that phrase, "could pass for real," so keep it in your head. If a viewer could reasonably think the image or video is a real photograph of a real thing, that is when your label kicks in.
I wrote a separate, plain walk-through of exactly how to label AI content so you can get the wording and placement right the first time.
You do not have to invent a label from scratch. The EU published a free, official set of icons for marking AI content. Download the official EU AI content labels (SVG and PNG), no attribution required, and drop the right one on your creative.
Does the EU AI Act Apply to US Brands?
Yes, and this is where a lot of US teams assume they are safe and are not. The rule follows where the ad is seen, not where your company is based (EU AI Act, Article 2, 2026). A brand in Chicago running an AI-built video ad that plays to audiences in Germany, France, or Spain is inside the regulation. If your media plan includes the EU and your creative includes AI-made people or scenes that could pass for real, the labeling job is yours.
What AI Content Needs a Label, and What Does Not?
Here is the short version, and there is a fuller checklist in does my ad need an AI label.
Ordinary cleanup does not need a label. Retouching a real photo, removing dust or blemishes, minor color and light adjustments, cropping, all of that is treated as normal editing and stays outside the rule. A real photo of a real thing is still a real photo.
Making something look real that is not does need a label. A fully AI-generated person in your ad, a fabricated scene presented as if it really happened, a real subject dropped into a fake but realistic location, those are the cases the law is built for. Intent does not save you here. Even if you were not trying to fool anyone, if the result looks authentic and it was made or changed with AI, label it (Greenberg Traurig, 2026).
The gray middle is where the "when in doubt, label it" rule earns its keep. A light AI background extension on a product shot probably reads as minor. A full swap of the background into a different real-looking place probably does not. When you cannot cleanly call it, the cheap and safe move is a label.
Is There a Break for Creative Work?
There is, but it is smaller than people hope. Genuinely artistic, satirical, or fictional work gets a lighter touch: you still note that AI was used, but you can do it quietly, in the credits or a title card, so it does not wreck the piece. The catch is that ordinary product advertising does not count as artistic work, so a normal commercial does not get the easy version (Greenberg Traurig, 2026). Do not plan a campaign around this exception.
What Are the EU AI Act Penalties?
The AI Act has fine tiers, and a labeling breach sits in the middle one: up to 15 million euros or 3% of total worldwide turnover, whichever is higher (EU AI Act, Article 99, 2026). Smaller companies are capped at the lower of the two figures. Each EU country enforces it, and the fines for the labeling rules are live from August 2, 2026.
For most large advertisers the fine is not the scary part. The scary part is the same one New York created: getting publicly caught passing off AI content as real once the law made the standard plain.
How Does This Compare to New York's Law?
Same direction, different size. New York's law, live since June 9, 2026, only covers AI-generated humans in ads, with penalties from 1,000 to 5,000 dollars per violation (New York Governor's Office, 2026). It is narrow by design. I broke that one down in New York's AI ad disclosure law.
The EU AI Act is the wide version: one rulebook across 27 countries, covering any AI-made image, audio, or video rather than only fake people, with fines measured against global turnover. California passed its own AI transparency law the same season, though that one leans on the AI tool makers rather than the advertiser. Even so, the direction is plain: within a few months, the EU, New York, and California all moved to put AI content on the record. The takeaway is not to build a process for each one. It is to build a single clean labeling habit that satisfies the strictest of them, the EU, and stop chasing the rest. If you want the case-by-case version, I broke it down in does my ad need an AI label.
What This Means If Your Work Is Mostly Real
Here is the good news if you make visual content the honest way. The line the EU drew sits close to where careful hybrid production already lives. The label attaches to AI content that could pass for real, and hits hardest on fake people and fabricated scenes. Photography of real food, real drinks, and real places, then extended or adapted with AI, is a different animal from a scene invented whole and passed off as real.
That is how we run hybrid production from the start. We photograph the real subject, with real light and real texture, then use AI to extend frames, build format variations, and turn one shoot into a season of assets. When there is a performance, it belongs to a person who was actually on set.
It also means your shoot records just became useful in a new way. Because these rules turn on what was made or changed with AI, being able to say exactly which assets were photographed, which were extended, and which were fully generated is worth having. We already keep that record for licensing, the same way we track usage rights on commercial shoots. Now it does double duty. If your team wants help building that habit into the way you work, that is what our AI consulting is for.
None of this is an argument against AI creative. For some campaigns it is the right call, and a clear label is a fair trade. It is an argument for knowing which tool you used and why, before the media plan locks.
How to Comply With the EU AI Act
Start with an audit, not a policy memo. Most teams do not actually know how much AI-made content is sitting in their live creative. A workable order:
- Find the AI in your campaigns that reach the EU. Flag every asset that was made or changed with AI, and mark anything that could be read as a real person, place, or event.
- Use tools that tag their output, and keep the tag. That hidden marker is part of your story, so favor tools that add it and do not strip it out in your edit and export.
- Write your label once. Settle on wording and placement that a normal viewer would clearly notice, then reuse it across formats so it fits the work instead of fighting it.
- Put a check in your approvals. One line at final review, "any AI content reaching the EU, and is the label there," costs nothing. Fixing ads after they ship costs a lot.
- When in doubt, label it. That single habit covers most of the gray area for a fraction of the risk.
If you are weighing how a campaign should split between shooting and generating, that is a conversation worth having before creative starts, not at legal review. Book a call and we will walk through how we structure hybrid campaigns that scale across markets without crossing the lines these rules just drew.
The AI labeling era is not coming. As of this summer it is here, in New York, in California, and across the EU. The brands that treat being upfront as part of the craft, rather than a constraint on it, are going to find this an easy standard to live with.
