The EU AI Act Is a 2026 Problem for Marketers, Not a 2027 One

The Digital Omnibus pushed the headline high-risk rules to December 2027. It did not touch Article 50. AI-content transparency obligations still apply from 2 August 2026 — and a surprising number of teams are planning around them as if the whole Act slipped a year. It didn’t.

The delay that wasn’t the delay you think

In mid-2026 the EU moved to simplify its own rulebook. The instrument is the Digital Omnibus, and its most-quoted effect is real: standalone high-risk AI obligations under Annex III moved from August 2026 to 2 December 2027, and product-embedded high-risk systems under Annex I moved to 2 August 2028 (Gibson Dunn, EU AI Act Omnibus — postponed high-risk deadlines). If you build or deploy a genuine high-risk AI system, that breathing room is a gift.

Most marketing teams do not build high-risk AI systems. They generate synthetic content — copy, images, video, chat replies — and that puts them squarely under Article 50, the transparency chapter. And Article 50 was not part of the delay. Its obligations remain on the original schedule and apply from 2 August 2026 (Sidley Data Matters, 24 June 2026, Preparing for compliance by 2 August 2026). So the headline that read “the AI Act got pushed to 2027” is, for a content team, actively misleading. The part that governs what you publish did not move.

One honest caveat on the status

The Omnibus is further along than the “provisional agreement” framing you may have read in spring. The European Parliament adopted it on 16 June 2026 and the Council gave its final green light on 29 June 2026 (Council of the EU, final green light, 29 June 2026). But as of 7 July 2026 it had not yet been published in the Official Journal, and it only enters into force on the third day after that publication. Until then, the original 2 August 2026 high-risk dates technically remain the legal baseline. Treat the high-risk delay as pending and expected — not final. If a decision of yours rests on it, re-check the Official Journal on the day. None of this touches Article 50, which stands either way.

What Article 50 actually asks of you

On 8 May 2026 the Commission’s AI Office published draft guidelines on Article 50 transparency (European Commission, consultation on draft Article 50 guidelines). They map to four obligation categories. Read them as four questions about your own content operation:

Interactive AI must disclose it is AI. If a chatbot or interactive assistant talks to a person, that person must be told they are interacting with an AI system.
Synthetic content must be machine-readable marked. Providers of systems that generate synthetic audio, image, video or text must mark the output so it can be detected as artificially generated.
Emotion-recognition and biometric categorisation must be disclosed. Deployers of those systems must inform the people exposed to them.
Deepfakes and AI-generated public-interest text must be disclosed. If you publish a deepfake, or AI-generated text on matters of public interest, the artificial nature must be disclosed (Greenberg Traurig, June 2026, Details of the transparency obligations).

Two details in the guidelines that catch teams out

Intent does not matter.The guidelines are explicit that whether content “falsely appears authentic or truthful does not depend on the deployer’s intention to deceive or mislead” (Greenberg Traurig; Covington Global Policy Watch, 10 Takeaways on the draft guidelines). You cannot argue your way out of a disclosure duty by saying you never meant to fool anyone.

A fictitious-but-realistic AI human still counts. A synthetic depiction of a person who does not exist can still be a deepfake “even where no identifiable rights-holder was implicated” — content qualifies if it resembles someone or something that can exist or could have existed in reality (Greenberg Traurig; Covington). The AI-generated brand spokesperson in your next campaign is in scope even though the face is invented.

A word on the test itself, because it is easy to state wrongly. It is nota raw “could any single viewer be fooled” test. The guidelines apply a broadened average-consumer standard that takes account of the actual, potentially diverse exposed audience — including foreseeable exposure to children, older viewers, or audiences with lower digital and AI literacy. In plain terms: whether a reasonable member of the actual audience, accounting for its less AI-literate members, could take the content as authentic.

The two dates that give you room to work

First, a grace window. The Art 50(2) marking duty carries a four-month grace to 2 December 2026 for generator systems that were already on the market before 2 August 2026 (Gibson Dunn; Sidley). This applies only to that marking duty — not to the broader Article 50 transparency obligations, which still land on 2 August.

Second, a soft-landing route. On 10 June 2026 the AI Office released the final Code of Practice on marking and labelling AI-generated content (Jones Day, June 2026, Final Code of Practice on marking and labelling; European Commission press release IP/26/1328). Signing gives a presumption of good-faith compliance and a smoother path to demonstrating compliance — it is not a safe harbour and not immunity from fines, so keep the expectation modest. The initial sign-up deadline to appear on the list before 2 August 2026 is 22 July 2026 (18:00 CEST).

The fine tiers, kept straight

Article 99 sets three tiers, and marketing copy routinely attaches the wrong one to labelling. The facts:

Up to EUR 35 million or 7% of worldwide annual turnover — Article 5 prohibited practices only. This is the number for the most serious breaches. It is not the number for a labelling failure.
Up to EUR 15 million or 3% of turnover — other operator obligations, including Article 50 transparency and high-risk duties. This is the tier that governs AI-content labelling.
Up to EUR 7.5 million or 1% — supplying incorrect or misleading information to authorities.

(Article 99, artificialintelligenceact.eu; EU AI Act Service Desk.) If you take one number away from this piece, make it this: labelling sits in the EUR 15M/3% tier. Any vendor waving EUR 35M at you over a missing content label is citing the wrong law.

And Article 5 has been live since 2025

Worth remembering, because it frames the manipulation question that governed content raises: the Article 5 prohibitions on manipulative and exploitative AI practices have applied since 2 February 2025. In May 2026 the Commission published three expert studies on Article 5, and the boundary between lawful personalisation and unlawful manipulation is genuinely still being worked out — personalised advertising is “not inherently manipulative” unless it deploys subliminal, manipulative or deceptive techniques (European Commission, three studies on Article 5). This is an evolving line, not a settled one — another reason to keep counsel close.

What a marketing team should actually do before August

None of this requires panic. It requires an inventory and a process. In order:

List every place your team ships AI-generated content into EU-facing communications — chat, copy, image, video, synthetic voices, AI spokespeople.
Decide your disclosure and marking approach for each, mapped to the four Article 50 categories above.
Keep a record that shows, per asset, what was AI-generated and how it was disclosed. When an authority or a customer asks, the answer should be a file, not a memory.
Decide on the Code of Practice before 22 July if you want the presumption of good-faith compliance from day one.
Get counsel to sign off on the edge cases — the human-editorial exception, the deepfake test, anything near Article 5.

The teams that treat 2 August 2026 as a real date, not a rescheduled one, will spend August publishing. The teams that believed the “2027” headline will spend it building the inventory they should have built in the spring.

Where are you on the 2 August timeline?

Take the readiness diagnostic.

A short, self-serve check of where your content operation sits against Article 50 transparency — an algorithmic diagnostic, not a legal assessment. Pair it with the marketing-team compliance guide when you want the detail behind each obligation.

Run the readiness diagnostic Read the compliance guide

Frequently asked questions

Did the Digital Omnibus delay Article 50 transparency to 2027?

No. This is the single most common misreading. The Omnibus postponed standalone high-risk obligations (Annex III) to 2 December 2027 and product-embedded high-risk (Annex I) to 2 August 2028. Article 50 transparency obligations were not part of that delay — they remain on the original schedule and apply from 2 August 2026 (Sidley Data Matters, 24 June 2026; Gibson Dunn). The only narrow carve-out is the Art 50(2) marking duty for generator systems already on the market before that date, which gets a four-month grace to 2 December 2026.

Is the Omnibus delay already legally binding?

Not yet, as of 7 July 2026. The Digital Omnibus was adopted by the European Parliament on 16 June 2026 and given final Council green light on 29 June 2026, but it had not been published in the Official Journal at the time of writing. It enters into force on the third day after that publication. Until then, the original 2 August 2026 high-risk dates technically remain the legal baseline. Treat the high-risk delay as pending and expected, not done — and re-check the Official Journal on the day you rely on it.

Which fine tier applies to AI-content labelling?

Article 50 transparency breaches sit in the up-to EUR 15 million or 3% of worldwide annual turnover tier (Article 99). The higher EUR 35 million or 7% tier is reserved for Article 5 prohibited practices only. Any marketing material that attaches the EUR 35M figure to a labelling failure is citing the wrong tier. Incorrect or misleading information supplied to authorities sits in a third tier, up to EUR 7.5 million or 1%.

Does the human-editorial exception get us out of the deepfake disclosure duty?

Only when it genuinely applies. The exception is narrow: it requires substantive review by a competent person with editorial responsibility for the publication — a real editorial process, not a skim or a rubber-stamp. Where a human takes editorial responsibility for AI-assisted output as part of an editorial process, the deepfake disclosure duty can be tempered. A workflow that merely routes content past someone does not meet the standard. This is a legal judgement to make with your counsel, not a checkbox.

Is this article legal advice?

No. It is a planning guide written for marketing and content teams from a fact-checked reading of public sources, current as of 7 July 2026. Regulatory dates and interpretations move — the Omnibus itself had not reached the Official Journal at the time of writing. Confirm your specific obligations with qualified counsel before making compliance decisions.

How Brand Memory Changes Your EU AI Act Exposure — mapping the four risk vectors to concrete mechanisms
EU AI Act for Marketing Teams — the Article 50 compliance pillar guide
The readiness diagnostic — a 90-second self-serve check
The EU AI Act whitepaper — the longer-form reference

Sources: Gibson Dunn (Omnibus & high-risk deadlines); Sidley Data Matters, 24 June 2026 (Article 50 by 2 August 2026); Council of the EU, 29 June 2026 (final green light); European Commission draft Article 50 guidelines, 8 May 2026, and three studies on Article 5, May 2026; Greenberg Traurig and Covington Global Policy Watch (deepfake & intent guidance); Jones Day and Commission IP/26/1328 (Code of Practice, 10 June 2026); artificialintelligenceact.eu and the EU AI Act Service Desk (Article 99 fine tiers). Current as of 7 July 2026 — verify the Official Journal status before relying on the high-risk delay.

The EU AI Act is a 2026 problem for marketers, not a 2027 one.