The EU AI Act Article 50 transparency rules have applied since 2 August 2026. For publishers, product teams and marketers, the main question is no longer when the rules start. It is which duties belong to them. Article 50 distinguishes between providers and deployers, with providers carrying a marking duty and deployers carrying a disclosure duty. The Commission's Code of Practice page adds that about 190 organisations had signed the voluntary code by the end of July.
Two roles and five duties
The Article 50 summary is short: providers must inform users when they are interacting directly with an AI system, and deployers must disclose deepfakes and AI-generated text. The detail is where the work lies.
There are two duties for providers. The first covers systems that talk to people. Users must be told they are dealing with an AI unless this is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect, given the circumstances and the context of use. That standard depends on context, and it does not excuse a support bot that looks like a human agent. The second duty covers generative systems that produce synthetic audio, image, video or text content. Providers of those systems must make sure the outputs are marked in a machine-readable format and detectable as artificially generated or manipulated.
Deployers carry the other duties. They must tell people exposed to an emotion recognition system or a biometric categorisation system that the system is operating. They must also disclose deepfakes and certain AI-generated text. The required format is spelled out too: information must be provided clearly, distinguishably and accessibly.
| Duty | Who carries it | What it requires | Key exception |
|---|---|---|---|
| AI talking directly to people | Provider | Tell users it's an AI | Obvious from context |
| Synthetic audio/image/video/text | Provider | Machine-readable, detectable mark | Law-enforcement use |
| Emotion recognition/biometrics | Deployer | Inform exposed people | Law-enforcement use |
| Deepfakes | Deployer | Disclose as AI-generated | Specific exceptions apply |
| Public-interest text | Deployer | Disclose as AI-generated | Human review + editorial responsibility |
Article 50 also exempts systems authorised by law to detect, prevent, investigate or prosecute criminal offences. For the chatbot duty, that exemption does not cover systems that are available for the public to report a criminal offence. That exemption is unlikely to matter for a marketing team.
When editing replaces the label
Publishers should read the text duty closely. Deployers of AI that produces text published with the purpose of informing the public on matters of public interest must disclose that the text was artificially generated or manipulated. The exemption applies where the content has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility.
Both conditions have to be met. A quick read-through by an editor is not enough if no person or legal entity holds editorial responsibility for the publication. A masthead with no human review behind it fails the test as well. Newsrooms that already run AI drafts through a named editor and a publisher of record will probably meet both conditions. Content farms that publish model output directly will not.
Deepfakes are a separate case. The Commission defines them as image, audio or video that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful. The editorial-review exemption is written into the text duty. It does not appear in the deepfake duty, so a publication that edits a synthetic video clip should not assume its newsroom workflow removes the need for a label. The EU has also created a set of icons that deployers of generative AI systems may use to label their AI-generated content.
Voluntary code, binding law
The Commission's position on the code is clear. Adherence is voluntary, but the transparency requirements under article 50 of the AI Act are legal obligations. Signing the code is optional. Following Article 50 is not.
Signing still offers an advantage. The Commission and the AI Board have confirmed that the code is an adequate voluntary tool for demonstrating compliance. The code was drawn up by independent experts in a multi-stakeholder process facilitated by the AI Office and the Code of Practice on transparency of AI-generated content has 2 sections. Section 1 covers providers and the marking and detection of content. Section 2 covers deployers and the labelling of deepfakes and AI-generated text. Uptake has been broad: by the end of July 2026, about 190 companies and organisations had signed the code.
The scope of the code matters for product teams. It helps providers and deployers of generative AI systems to comply with the AI Act's obligations for labelling and marking of AI-generated content – Article 50(2), (4) and (5) of the AI Act. It does not address the duty to tell users they are talking to an AI, or the duty around emotion recognition. Under the law itself, the Commission's adequacy assessment covers the obligations laid down in paragraphs 2 and 4. A company that signed the code and also runs a customer-facing chatbot still has to design that chatbot's disclosure on its own.
The law sets a demanding technical standard for providers' marking. Solutions must be effective, interoperable, robust and reliable as far as this is technically feasible, taking into account content type, implementation cost and the state of the art. If the Commission finds the code inadequate, it may adopt an implementing act specifying common rules instead.
What this means for each team
Marketers and publishers mostly act as deployers when they use AI. That means the machine-readable marking duty generally falls on the companies that build the image, video and text tools they rely on. Their own responsibilities are the visible ones: labelling deepfakes, and either disclosing AI-generated public-interest text or putting it through documented human review under a responsible editor.
Product teams may hold both roles. Building a generative feature makes a company a provider, with marking obligations. Adding a chatbot to an app brings the AI disclosure duty, and the code offers no compliance route for that duty.
Signing the code is worth considering because it gives a recognised way to show compliance with the marking and labelling duties. It does not satisfy every Article 50 obligation. The Commission's Code of Practice page explains how to sign, through the Signatory Form and instructions.
Sources
- ai-act-service-desk.ec.europa.eu – "AI Act Service Desk - Article 50: Transparency obligations for providers and deployers of certain AI systems"
- digital-strategy.ec.europa.eu – "Code of Practice on Transparency of AI-generated Content"