Skip to content
Brands & Campaigns August 6, 2026

Your AI Ad Needs a Label in Europe Now

Article 50 started applying on Sunday. Brands and agencies both count as deployers, both carry the labeling duty, and the tool vendors got four extra months that you did not.

By The State of AI Marketing newsroom
Share
Editorial illustration for: Your AI Ad Needs a Label in Europe Now
Credit: JAC Growth Marketing

Article 50 of the EU AI Act started applying on Sunday, 2 August. For anyone running ads that reach people in the European Union, two obligations now sit on the brand rather than on the software.

One covers realistic synthetic media. A deployer using AI to generate or manipulate image, audio or video that constitutes a deep fake has to disclose it as artificially created or manipulated. The other covers text published to inform the public on matters of public interest. That has to be disclosed as AI-generated, unless a person took editorial responsibility for it after substantive human review.

Both hinge on one word. “Deployer” means you.

Four lawyers at Lewis Silkin, including advertising and marketing co-heads Brinsley Dresden and Geraint Lloyd-Taylor, published a read of the rules two days before they took effect. Their conclusion is the part worth pinning to a wall. Being a deployer doesn’t require touching the software. It’s enough that a party decides for what purposes and how an AI system gets used, which puts the brand and the agency in the same position at once:

“Both the advertiser and the agency can be ‘deployers’ in the same chain, each in a different respect”

They go further on what that does to the usual contractual reflex, noting that “neither party has exclusive control, yet both can hold ‘authority’, meaning compliance is very likely a joint responsibility that cannot be quietly left to the other party.” Their recommendation is to agree early in the creative process what label to use, where it appears, and how to allocate labelling responsibility and risk contractually.

Joint responsibility is the detail that turns this from a legal update into a workflow problem. There’s no version where the agency absorbs it and the brand finds out later.

That pattern is becoming familiar. When Google began generating targeting, ad copy and landing pages on the advertiser’s behalf, the liability stayed with the advertiser. Article 50 does the same thing to synthetic creative: the machine makes it, and the brand answers for it.

Now the asymmetry. Article 50 also requires machine-readable marking, the invisible provenance signal that lets a platform detect synthetic content automatically. That duty falls on providers, meaning the companies that build the generative tools. Cooley’s Brussels team noted in an alert on Monday that providers whose systems were already on the market have until 2 December 2026 to comply. That grace period came in with the Digital Omnibus package, which entered into force in late July.

Read those two dates together. The tool you use to make the ad has four more months to build the technical marking. You had until Sunday to put a visible label on the ad. The party with the engineering capacity got the extension, and the party holding the brand risk did not.

Penalties for getting Article 50 wrong run to €15M or 3% of worldwide annual turnover, whichever is higher. Reach is the other thing to register. The rules apply to output used in the EU, so a US brand with European customers is in scope whether or not it has a European entity.

The industry has been asking for guidance on this for months. Gabrielle Robitaille, director of policy and AI community lead at the World Federation of Advertisers, described the inbound to Digiday back in April, when legal teams started pushing the question back at marketers:

“Over the past couple of months alone, I must have had a good five, six or seven brands reach out asking, ‘do we have guidance?’ and ‘do we have insights into how brands are thinking about this issue?’”

Her account matters because of where it locates the confusion. The brands calling her weren’t asking whether they could avoid disclosure. They were asking what disclosure looks like.

Four months later the clarity arrived as a date, and the practical questions Robitaille was fielding are still open. Article 50 says a deployer must disclose. It doesn’t hand you a badge, a placement, or a wording standard.

That vacuum is where the cost sits. Stamping “AI-generated” across a hero image is a creative decision as much as a compliance one, and it gets made without knowing what competitors will do or how the label reads to a customer. The law tells you to speak. It doesn’t tell you how loud.

There’s a narrower reading worth holding onto before anyone relabels a whole library. Lewis Silkin’s team notes that routine advertising copy generally falls outside the public-interest text requirement. That obligation is aimed at something closer to a press release or a crisis statement than a product description. The deep fake rule is the one that reaches ordinary campaign work, and it turns on realism: synthetic humans, AI-generated product shots, manufactured scenes a viewer would take as real.

Which sets a practical boundary. Recolouring a background is assistive editing. Generating a person who does not exist and putting them in a testimonial is not.

The timing is awkward for anyone mid-pilot. Conair ran an AI-generated Cuisinart video against one its own team made, and the AI version came out ahead. That’s the exact category of asset the deep fake rule reaches once the footage looks real. A test that was a creative question in July is a disclosure question in August.

Three moves follow, and none of them require a legal opinion first.

Decide your label now, as a design decision, before a campaign forces it. Pick the wording, the placement and the minimum size, and put it in the brand guidelines next to the logo rules, because the alternative is deciding it under deadline on the asset that matters most.

Then open the contracts. Both parties can be deployers, so the MSA needs a clause that names who applies the label, who keeps the record of what was generated, and who pays if it’s wrong. An agency agreement written before August 2026 almost certainly says nothing about this.

Then audit what is already live. Anything realistic and synthetic that’s running to EU audiences today is running under the new rules, not the old ones, and the four-month grace period belongs to your tool vendor rather than to you.

Quoted in this story

  • Gabrielle Robitaille, Director of Policy and AI Community Lead, World Federation of Advertisers (source)
  • Brinsley Dresden, Partner and Co-Head of Advertising & Marketing, Lewis Silkin (source)

Want your perspective in coverage like this? Get quoted.

Sources

This story is part of our running coverage: the full picture →

Get Net Effect.

The net effect of AI on your marketing: the stories that matter, twice a week, in five minutes.

More from Brands & Campaigns