Regulation (EU) 2024/1689 · Article 50
The rules don't ask “did AI write this?” They ask who signed off.
In force 2 August 2026
The answer in 30 seconds
- →From 2 August 2026, the EU AI Act's transparency rules (Article 50) apply. Breaking them can cost up to €15 million or 3% of worldwide annual turnover.
- →Chatbots must tell people they are talking to AI. AI-generated images, video and audio must carry a machine-readable mark. Realistic synthetic people, voices or events need a visible label.
- →AI-drafted text does not need a label when a human has reviewed it and a named person or company holds editorial responsibility.
- →So the working question is not “did AI write this?” — it is “who approved it, and can you prove it?”
Does this apply to your team?
Almost certainly. Article 50 of the EU AI Act — Regulation (EU) 2024/1689 — is not written for AI companies; it is written for everyone who uses AI in front of other people. If your team runs a chatbot, generates campaign visuals, drafts posts or emails with AI, or publishes AI-assisted articles, you fall under at least one of its four duties from 2 August 2026.
The law splits responsibility between two roles. The provider builds or supplies the AI system — that is your vendor, or you, if you white-label a model under your own brand. The deployer uses the system professionally — that is your marketing team. Two of the four duties below sit with providers, two sit with you. You need to know both: a provider's failure becomes your problem the moment their unmarked output ships in your campaign.
The four duties, translated to marketing
Art. 50(1) — your chatbot must say it's a bot
Regulation (EU) 2024/1689, Article 50(1)
“Providers shall ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system…”
In practice: the disclosure must be there at first contact, not buried in terms. The exemption — “unless obvious to a reasonably well-informed person” — is narrower than it sounds; a friendly first-name persona pushes you away from it, not toward it.
Do — “You're chatting with Tisser's AI assistant. A colleague reads along when needed.” One line, in the chat window, before the first answer.
Art. 50(2) — AI-generated media must be machine-readably marked
Regulation (EU) 2024/1689, Article 50(2)
“Providers of AI systems… generating synthetic audio, image, video or text content, shall ensure that the outputs… are marked in a machine-readable format and detectable as artificially generated or manipulated.”
In practice: this one sits with your tooling vendors — but it lands on your desk twice. First, ask every AI vendor in your stack how they mark output (metadata, watermark) and get the answer in writing. Second, don't strip the mark: exporting through a tool that wipes metadata undoes your vendor's compliance in your name.
Ask — one line for your vendor review: “How do your outputs satisfy Article 50(2) marking, and does the mark survive your export formats?”
Art. 50(4) — realistic synthetic people, voices and events need a visible label
Regulation (EU) 2024/1689, Article 50(4), first subparagraph
“Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake, shall disclose that the content has been artificially generated or manipulated.”
In practice: “deepfake” is broader in the Act than in the news. A synthetic spokesperson in your product video, a cloned voice-over, an AI image that looks like a real photo of a real place — all of it needs a disclosure the viewer actually sees. Stylised illustration and obviously artificial imagery are outside this duty; photorealism is the trigger.
Do — “Presenter generated with AI” as an on-screen caption, not a footnote on a landing page the viewer never opens.
Art. 50(4) — AI-written text: label it, or have a human take responsibility
Regulation (EU) 2024/1689, Article 50(4), second subparagraph
“…deployers of an AI system that generates or manipulates text which is published with the purpose of informing the public on matters of public interest shall disclose that the text has been artificially generated or manipulated. This 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…”
In practice: this is the paragraph that should shape your workflow. Plain product copy rarely counts as “informing the public on matters of public interest” — but thought leadership, industry commentary and news-style content can. Instead of debating each piece, use the exemption the law hands you: route every AI draft through named human review before publication. Reviewed and signed off, the labeling duty for text falls away — and you hold evidence of the review.
Note — this is why we built the publish gate: AI drafts, a named person approves, the approval is logged. The exemption is only worth something if you can show the review happened.
Where 2 August 2026 sits in the rollout
- 1 Aug 2024
- The AI Act enters into force. The clock starts.
- 2 Feb 2025
- Prohibited practices banned (manipulative techniques, social scoring). AI-literacy duty for staff begins.
- 2 Aug 2025
- Obligations for general-purpose model providers (your vendors) apply.
- 2 Aug 2026
- Article 50 transparency duties apply — everything on this page. Most of the rest of the Act applies too, and member-state enforcement must be in place.
- 2 Aug 2027
- Final deadline for high-risk AI embedded in regulated products.
What non-compliance costs
Breaching Article 50 can be fined up to €15 million or 3% of total worldwide annual turnover, whichever is higher (for SMEs: whichever is lower). Enforcement runs through national market surveillance authorities; in the Netherlands the Autoriteit Persoonsgegevens is preparing the coordinating role.
Regulation (EU) 2024/1689, Article 99(4) and 99(6)
The before-August checklist
- 01Inventory every AI touchpoint — chatbot, image and video generation, drafting tools, voice. If your count surprises you, that's normal: organizations average 67 AI tools (Prompt Security, 2025).
- 02Add the chatbot disclosure line at first contact, in every language you operate in.
- 03Get vendor marking confirmed in writing for every generative tool, and test that marks survive your export pipeline.
- 04Set a house rule for photorealistic synthetic media: visible label, on the asset itself, no exceptions.
- 05Route all AI-drafted text through named human review and log who approved what, when.
- 06Brief the team once — the four duties fit on one slide, and the AI-literacy duty (Article 4) already applies to you anyway.
Questions your team will ask
Does a blog post written with AI need an “AI-generated” label?
Not if a human reviewed it and a named person or company holds editorial responsibility — that is the explicit exemption in Article 50(4). Without that review process, a label is required when the text informs the public on matters of public interest. The review must be real and demonstrable, not a rubber stamp.
Do AI-edited photos count as deepfakes?
Routine retouching does not make content a deepfake. The duty triggers when content appears to be an authentic depiction of real people, places or events while it is generated or manipulated — a synthetic presenter, a cloned voice, a fabricated “photo”. When in doubt, label: a caption costs nothing.
Our chatbot vendor says they're compliant. Are we covered?
Partly. The design duty in 50(1) sits with the provider, but you choose how the bot is presented on your channels. Hiding or undoing the disclosure — or stripping marks from generated media in your pipeline — creates your own exposure as deployer.
Is there official guidance on how exactly to label?
The European Commission is finalising guidelines and a Code of Practice on marking and labeling AI-generated content. Until they land, the safe pattern is the one in this article: disclose at first contact, label on the asset itself, and keep review logs. We update this page when the guidance is published.
Sources
- Regulation (EU) 2024/1689 (EU AI Act), Articles 4, 50, 99 and 113 — EUR-Lex, consolidated text.
- European Commission, AI Act implementation timeline and Article 50 guidance workstream (2025–2026).
- Prompt Security, State of Shadow AI, 2025 — survey across organizations.
This article explains the rules in plain language; it is not legal advice. Reviewed 23 Jul 2026 · we update within a month of regulatory changes.
The Tisser position on Article 50