From 2 August 2026, any organisation that uses AI to generate or manipulate text that is published to inform the public on matters of public interest must disclose that the text is artificially generated or manipulated. This follows from Article 50(4) of the AI Act, the second limb. The duty sits with the deployer and the scope is narrow: not all AI text falls under it, only text on matters of public interest. Text is explicitly not a deepfake; the deepfake duty applies to image, audio and video and is a separate obligation.
This page accompanies our practical overview of Article 50. Below you will find what this duty entails, which text it covers, who carries it and when the important exception applies.
Which text falls under this duty?
The duty applies to AI-generated or manipulated text that is published to inform the public on matters of public interest. The scope is narrow. It does not concern all AI text, but text on matters of public interest, such as journalism-style publications. Ordinary commercial text, such as product descriptions or internal documents, generally falls outside it.
Important: text is not a deepfake. An AI-written article is something other than a fabricated video. The deepfake duty applies to image, audio and video and is a separate obligation. Do not conflate the two.
Who must disclose AI text?
The duty sits with the deployer: the organisation that uses the AI system under its own responsibility to generate or manipulate the text and that publishes the text to inform the public.
| Question | Answer |
|---|---|
| Who carries the duty? | The deployer who generates or manipulates and publishes the text |
| For which text? | Text that informs the public on matters of public interest |
| From when? | 2 August 2026 |
| Does it apply to all AI text? | No, only to text on matters of public interest |
The misconception that all AI text must be labelled is exactly where organisations go wrong. A product description or an internal memo generally falls outside the duty. An AI-written publication that informs the public on a societal matter falls under it.
How and when must the disclosure be made?
The disclosure must be clear and given at publication. Anyone publishing text to inform the public on matters of public interest discloses at the moment of publication that the text is artificially generated or manipulated.
The duty falls away when the AI-generated content has undergone a process of human review or editorial control and a natural or legal person holds editorial responsibility for the publication. The review must be substantive, not superficial. Anyone who merely glances over the text and hits publish does not meet this condition.
When does the exception apply?
This is the core of the second limb. The disclosure duty does not apply when two conditions are met together: the AI-generated content has undergone a process of human review or editorial control, and a natural or legal person holds editorial responsibility for the publication. The review must be substantive.
This exception is powerful, but it requires discipline. Anyone relying on it must be able to explain that the review was more than superficial and that an identifiable party holds editorial responsibility. Without those two elements, the disclosure duty simply continues to apply.
Text versus deepfake: two separate rules
Two duties are often confused. The disclosure duty for AI text under Article 50(4), second limb, only applies to text that informs the public on matters of public interest, and falls away with substantive editorial review by a responsible party. The deepfake duty under Article 50(4) applies to image, audio and video that appear authentic and is a separate obligation. Anyone publishing both text and image with AI may face both.
What should you do now?
Map AI-generated public-interest text
Map where in your organisation AI is used to generate or edit text that is published to inform the public on matters of public interest.
Test role and editorial review
Determine per publication whether you are the deployer and whether the text falls under the scope. Establish whether a human carries out substantive editorial review and holds editorial responsibility.
Build the disclosure into the process
Build the disclosure into your publication process for the cases where the exception does not apply. Keep evidence of the editorial review and of who holds responsibility.
Organisations that arrange this before the summer will not have to improvise on 2 August 2026.
Frequently asked questions about labelling AI text
Practical questions about the disclosure duty for AI-generated text under Article 50(4) of the AI Act.