Regulation (EU) 2024/1689, the EU AI Act, applies from 2 August 2026. Article 113 states it plainly, with a short list of provisions that started earlier or start later. Article 50, which carries the transparency obligations, is not on that list, so it applies now.
The summaries circulating since then mostly land on one sentence: AI content must be labelled. The regulation itself says something more specific, and the specifics decide whether this touches your marketing team at all.
This is a description of what the text says. It is not legal advice, and anyone with real exposure should get their own.
What Article 50 sets out
The article has four operative obligations, and they are aimed at different people.
Paragraph 1 is about disclosure of interaction. Providers must design AI systems that interact directly with people so that people are informed they are dealing with an AI system, unless that is obvious to a reasonably well-informed and observant person given the context.
Paragraph 2 is about machine-readable marking, and it lands on providers of AI systems, including general-purpose systems, that generate synthetic audio, image, video or text. Their outputs must be “marked in a machine-readable format and detectable as artificially generated or manipulated.” This is an obligation on whoever builds and supplies the model, not on the marketer using it. The text also carves out systems performing “an assistive function for standard editing” or those that do not substantially alter the input or its meaning.
Paragraph 3 covers emotion recognition and biometric categorisation, with a duty to inform the people exposed to it.
Paragraph 4 is the one relevant to publishing, and it splits in two. For image, audio or video content constituting a deep fake, deployers must disclose that the content is artificially generated or manipulated. For text, the wording is narrower: the duty applies to text “which is published with the purpose of informing the public on matters of public interest.”
The exemption in the second half of paragraph 4
Read paragraph 4 to the end. The text disclosure obligation “shall not apply … where the AI-generated content has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication of the content.”
Both conditions have to hold. Someone reviewed it, and someone is accountable for publishing it.
That is a meaningful exemption for anyone running an editorial process, and it is the part that disappears when Article 50 gets compressed into a blanket labelling mandate. It also means the compliance question for a publisher is less “did AI touch this” and more “can we show a human reviewed it and that responsibility sits somewhere.”
There is a second limit worth noticing. Even without the exemption, the text obligation attaches to publishing on matters of public interest. A product page, a pricing FAQ, or a case study is not obviously that. A commentary piece on an election, a public health claim, or a contested news topic plausibly is. The line is not sharply drawn in the text, which is exactly why the answer for a given page belongs to a lawyer rather than a blog post.
What the July amendment changed
A summary written from the 2024 text is already out of date. The regulation was amended six weeks ago, by Regulation (EU) 2026/1744 of 8 July 2026, published in the Official Journal on 24 July 2026. Three points from it matter here.
Article 50 paragraphs 1 to 6, the operative transparency duties quoted above, are untouched. They apply as written.
There is now a transitional period, and it is a narrow one. The amendment inserts a new paragraph into Article 111 stating that providers of systems generating synthetic audio, image, video or text content “that have been placed on the market before 2 August 2026 shall take the necessary steps in order to comply with Article 50(2) by 2 December 2026.” The recitals describe this as a four-month transition.
Read the scope carefully, because it is easy to inflate. This defers one paragraph, paragraph 2, for one group, providers whose systems were already on the market, and it does not touch anything else in Article 50. The disclosure duties in paragraphs 1, 3 and 4, including the deep fake and public-interest text rules, apply now regardless of when a system reached the market. December is not a grace period for Article 50. It is a deadline for retrofitting machine-readable marking into products that predate the regulation’s application date.
The high-risk timetable moved as well, which is the part that generated headlines. Article 113 now sets those obligations to apply from 2 December 2027 for systems classified as high-risk under Annex III, and from 2 August 2028 for those under Annex I. Note what this does not do: it does not delay Article 50. The transparency obligations were never on that timetable.
One smaller change. Article 50(7) previously let the Commission approve codes of practice through implementing acts. The amendment removes that empowerment, leaving the Commission to assess whether adherence to a code is adequate, taking account of the Board’s opinion. The recitals reason that these codes have limited legal effect and do not confer a presumption of conformity, so formal approval was not necessary.
We are not repeating any penalty figures. Fine levels get quoted confidently in secondary coverage, frequently without a link to the article they come from. If a number matters to your decision, it needs to come from the regulation text or from counsel, not from a listicle.
What to do with this
If you publish text on public-interest topics, find out whether your process produces evidence of human review and clear editorial responsibility, because those are the terms the exemption uses.
If you publish deep-fake-adjacent visual content, the disclosure duty in paragraph 4 has no equivalent editorial carve-out, and the artistic or satirical exception only limits how the disclosure is made rather than removing it.
If you are a marketing team producing ordinary commercial copy with an AI assistant, the paragraph 2 marking obligation is not yours. It sits with the provider of the system you are using, and for systems already on the market that provider has until 2 December 2026.
And if any of this is close to the line for your organisation, read Article 50 in the Official Journal rather than a summary of it. It runs to seven paragraphs and takes a few minutes.
Sources
- Regulation (EU) 2024/1689 (Artificial Intelligence Act), Official Journal: https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=OJ:L_202401689 (retrieved 23 August 2026). Source of all quoted wording from Article 50 and of the original application dates in Article 113.
- Regulation (EU) 2026/1744 of 8 July 2026, amending Regulation (EU) 2024/1689, Official Journal of 24 July 2026: https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=OJ:L_202601744 (retrieved 23 August 2026). Source of the new Article 111 transitional provision for Article 50(2), the revised Article 113 application dates for high-risk systems, and the amendment to Article 50(7). All three were read in the enacting terms, not in the recitals.
