Article 50 is the EU AI Act’s main transparency provision for certain AI systems. It requires notices when people interact with some AI systems, machine-readable marking of synthetic outputs by providers, and disclosures for emotion recognition, biometric categorisation, deepfakes and certain AI-generated public-interest texts. The precise duty depends on whether an organisation is the provider or the deployer, what the system does and how the output is used.
The Article 50 duties become applicable on 2 August 2026. The European Commission published a voluntary Transparency Code of Practice on 10 June 2026, and the Commission and AI Board assessed it as adequate on 8 and 9 July. The Commission says its separate interpretive guidelines are still being finalised for publication before 2 August 2026.
Article 50 at a glance
Different duties apply depending on an organisation’s role.
Not every AI-assisted task or publication is covered.
A limited transitional period applies to certain pre-existing systems.
Information must be given no later than first interaction or exposure.
Who must comply: provider or deployer?
The AI Act separates the organisation that places an AI system on the market from the organisation that uses it under its authority. The same company can be a provider in one situation and a deployer in another.
| Role | Plain-language meaning | Typical Article 50 duties |
|---|---|---|
| Provider | Develops an AI system, or has it developed, and places it on the market or puts it into service under its own name or trademark. | Interaction notice and machine-readable marking. |
| Deployer | Uses an AI system under its authority for professional or organisational purposes. | Notices for emotion recognition or biometric categorisation and disclosures for deepfakes or certain public-interest text. |
Legal definitions: AI Act Article 3(3) and 3(4).
What obligations does Article 50 contain?
Tell people when they are interacting with an AI system
Providers of AI systems intended to interact directly with people must design them so that users are informed they are interacting with AI, unless this is obvious to a reasonably well-informed, observant and circumspect person in the circumstances.
Make synthetic outputs machine-readable
Providers of systems that generate synthetic audio, images, video or text must ensure outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. The technical solution must be effective, interoperable, robust and reliable as far as technically feasible.
The duty does not apply to the extent that a system only performs standard assistive editing or does not substantially alter the input data or its meaning.
Inform people exposed to emotion recognition or biometric categorisation
Deployers must inform people when an emotion-recognition or biometric-categorisation system is operating on them. Applicable EU data-protection rules remain relevant.
Disclose deepfakes
Deployers using AI to generate or manipulate image, audio or video content that constitutes a deepfake must disclose that the content was artificially generated or manipulated.
The AI Act defines a deepfake as AI-generated or manipulated image, audio or video that resembles existing persons, objects, places, entities or events and would falsely appear authentic or truthful. Evidently artistic, creative, satirical, fictional or analogous works still require disclosure, but it may be presented in a way that does not hamper the work’s display or enjoyment.
Disclose certain AI-generated public-interest text
Deployers must disclose AI-generated or manipulated text published for the purpose of informing the public on matters of public interest.
This text-specific duty does not apply where the content has undergone human review or editorial control and a natural or legal person holds editorial responsibility for its publication. That exception is written for this category of text; it is not a general exemption for deepfakes.
Give the information clearly and at the right time
The notices and disclosures in paragraphs 1–4 must be clear, distinguishable and accessible, and provided no later than the first interaction or exposure.
Does every AI-generated item need a label?
No. Article 50 is not a universal rule saying “AI was used, therefore add a visible label.” The assessment depends on several questions:
- Is the organisation acting as provider or deployer?
- Is the system directly interacting with people?
- Is the obligation technical marking or visible disclosure?
- Is the output text, image, audio or video?
- Does image, audio or video meet the legal definition of a deepfake?
- Was text published to inform the public on a matter of public interest?
- Was there meaningful human review or editorial control, and who holds editorial responsibility?
- Does a specific exception apply?
A grammar correction, an internal draft and a realistic AI-generated video of a public figure do not raise the same Article 50 question.
Simple examples
| Situation | Likely Article 50 issue |
|---|---|
| A customer-service chatbot speaks directly to website visitors. | Provider-side interaction notice may be required unless the AI nature is obvious. |
| A communications team uses AI only to correct spelling in a press release. | The provider’s machine-readable marking duty may fall within the standard-editing limitation; visible deployer disclosure is not automatically required. |
| An agency publishes a realistic AI video showing a real politician saying words they never said. | Deepfake disclosure assessment is required. |
| An organisation publishes an AI-generated explainer about an election without substantive human review. | The public-interest text disclosure duty may be relevant. |
| An editor substantively checks, changes and approves an AI-assisted policy article, and the publisher takes editorial responsibility. | The text-specific human-review and editorial-responsibility exception may be relevant; the facts should be documented. |
These examples are general illustrations, not definitive legal assessments.
How does the Code of Practice relate to Article 50?
The law and the Code are not the same thing. Article 50 is binding when an organisation falls within its scope. Signing and following the Code of Practice on Transparency of AI-Generated Content is voluntary.
The Code provides practical measures for Article 50(2), (4) and (5): technical marking and detection for providers, and labelling of deepfakes and certain AI-generated text for deployers. The Commission and AI Board have found it adequate as a voluntary tool for demonstrating compliance. Organisations that use a different method remain responsible for showing that their measures meet Article 50.
Creates the legal transparency duties.
Provides an EU-recognised implementation framework.
Will clarify scope and application; final publication was pending on the review date.
When does Article 50 apply?
The Article 50 transparency requirements become applicable on 2 August 2026. According to the Commission’s FAQ, AI systems placed on the market before that date benefit from a transitional period for compliance with Article 50(2) and (4) until 2 December 2026.
Because guidance is still being completed close to the application date, organisations should record which source and version they relied on when designing their process.
What should an organisation do first?
- Map roles. Determine where the organisation is a deployer and whether it could also be a provider.
- List recurring use cases. Separate chatbots, text generation, image generation, audio, video, biometric categorisation and emotion recognition.
- Identify publication contexts. Distinguish internal content from public content and ordinary commercial content from matters of public interest.
- Define review authority. Record who can substantively change, reject and approve content and who holds editorial responsibility.
- Create a repeatable record. Capture the AI tool, content type, reviewer, decision, rationale, date and published version without collecting unnecessary confidential material.
- Escalate uncertainty. Treat unclear deepfake, public-interest and editorial-responsibility questions as legal-review points rather than automated conclusions.
Understanding the rule is only the first step.
ShowAI maps how AI content is created, reviewed, approved and documented, then converts recurring decisions into a workable internal process.
Primary sources
- Regulation (EU) 2024/1689 — official AI Act text on EUR-LexArticle 3 definitions and Article 50 legal duties.
- AI Act Service Desk — Article 50Official EU navigation and consolidated explanation of the provision.
- European Commission — Code of Practice on Transparency of AI-Generated ContentCode scope, voluntary status and adequacy assessment.
- European Commission — Code of Practice FAQApplication dates, transitional period and status of the guidelines.
- European Commission opinion on the Code’s adequacyCommission assessment dated 8 July 2026 and AI Board assessment noted for 9 July 2026.
This guide prioritises the binding regulation and official EU implementation sources. It distinguishes the legal text, the voluntary Code and practical recommendations. It does not constitute legal advice.