Background
The transparency obligations under Article 50 of the European Union’s AI Act apply from 2 August 2026. Contrary to a widespread assumption, businesses will not have to label every text, image or video created with the assistance of artificial intelligence.
Article 50 instead covers specific situations expressly defined by the Regulation. The applicable requirements depend in particular on whether a business acts as a provider or a deployer of an AI system. The obligations are not limited to high-risk AI systems.
No blanket duty: The role is decisive
Article 50(1) and (2) of the EU AI Act applies to providers of AI systems. Providers of systems intended to interact directly with people – such as chatbots, virtual assistants or automated telephone systems – must generally design those systems so that the persons concerned are informed that they are interacting with AI.
Providers of generative AI systems must also ensure that artificially generated or manipulated audio, image, video and text content is marked in a machine-readable format and can be detected as such. This technical provider obligation is intended in particular to facilitate the traceability and automated detection of synthetic content.
An exception may apply, among other situations, where a system merely performs assistive standard editing or does not substantially alter the input data or its meaning. A grammar check must therefore be assessed differently from the independent generation of a new text.
Businesses using existing AI systems for their own purposes will generally be deployers. Article 50(3) and (4) is particularly relevant to them. Article 50(3) concerns the use of emotion-recognition and biometric-categorisation systems. The natural persons exposed to such systems must be informed; data protection requirements in particular continue to apply alongside these obligations.
Deepfakes: Disclosure by deployers
Under Article 50(4) of the EU AI Act, deployers must disclose when they use an AI system to generate or manipulate a deepfake.
According to Article 3(60) of the EU AI Act, a deepfake is AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful. This may include realistic depictions of events that never occurred or artificially generated recordings that create the impression of a real product, person or situation.
According to the draft guidelines described in the attached source, clearly fantastical or physically impossible depictions should generally not be classified as deepfakes. Special rules also apply to artistic, creative, satirical, fictional or comparable works: disclosure may be made in an appropriate manner that does not hamper the display or enjoyment of the work.
AI-generated text on matters of public interest
Article 50(4) of the EU AI Act does not establish a general labelling obligation for text either. Disclosure is required in particular where a deployer publishes AI-generated or manipulated text for the purpose of informing the public on matters of public interest.
The decisive factor is therefore not merely whether an AI tool was used to create the text. The purpose and context of the publication are also relevant.
Timing and form of the information
Information required under Article 50 of the EU AI Act must generally be provided no later than the time of the first interaction or first exposure to the relevant content. For a chatbot, this may mean before or at the beginning of the conversation. For audio or video content, a notice may be required at the beginning of the contribution.
The information must be clear, distinguishable and accessible. Notices hidden in general terms and conditions, barely legible labels on images or information displayed for only a very short time are unlikely to meet these requirements.
The attached guide was published on 14 May 2026 and refers to the draft Commission guidelines and the second draft of a code of practice for AI-generated content available at that time. The final status of these supplementary documents should be considered separately when implementing the requirements.
What businesses should prepare
An undifferentiated internal rule requiring every AI-assisted item of content to be labelled does not accurately reflect the structure of Article 50 of the EU AI Act. A documented assessment process distinguishing between the systems used, the business’s roles and the different forms of publication is more appropriate.
This may include an inventory of the AI systems used, clearly defined review and approval processes, documented editorial responsibilities and comprehensible internal AI policies. Training can additionally raise employees’ awareness of deepfakes, publications concerning matters of public interest and other relevant AI compliance requirements.
To the point
- The transparency obligations under Article 50 of the EU AI Act apply from 2 August 2026.
- There is no blanket labelling obligation for all content created or edited with AI.
- Providers and deployers of AI systems are subject to different obligations.
- Deployers must assess potential deepfakes and certain AI-generated texts concerning matters of public interest in particular.
- Human review and editorial responsibility may provide a statutory exception for affected text content.
- Businesses should introduce role- and use-case-specific AI processes rather than an undifferentiated labelling rule.
Source: https://artificialintelligenceact.eu/de/transparency-rules-article-50/