What is the AI labeling obligation?
The AI labeling obligation brings together the transparency obligations in Article 50(1) to (4) of Regulation (EU) 2024/1689 (the AI Act). People should be able to tell that they are talking to an AI or that content was artificially generated or manipulated. The law requires two different actions for this:
- Marking is technical: the provider of a generative system marks its outputs in a machine-readable format (paragraph 2).
- Disclosure is aimed at people: they learn that they are dealing with an AI system or AI-generated content (paragraphs 1, 3 and 4).
How Article 50 fits into the other obligations of the regulation is shown in the EU AI Act guide. The scope is set out in detail in the Commission guidelines on the transparency obligations under Article 50.
Who must label what? The obligations in Article 50(1) to (4)
Article 50 of the AI Act assigns each obligation to a role. Under Article 3(3), a provider is whoever develops an AI system or has it developed and places it on the market under its own name. Under Article 3(4), a deployer is whoever uses it under its own authority, except in a personal, non-professional activity. The article on roles and obligations under the AI Act explains which role your company has for each system.
| Paragraph | Who | Obligation | Main exception |
|---|---|---|---|
| Para. 1 | Provider | design systems intended to interact directly with people, such as chatbots or voice assistants, so that users are informed they are interacting with an AI | The AI interaction is obvious to a reasonably well-informed, observant and circumspect person. |
| Para. 2 | Provider | mark synthetic audio, image, video and text outputs in a machine-readable format and make them detectable as artificially generated or manipulated | assistive function for standard editing, or no substantial alteration of the input data |
| Para. 3 | Deployer | inform people exposed to an emotion recognition or biometric categorization system of its operation | law enforcement permitted by law, with safeguards |
| Para. 4, first subparagraph | Deployer | disclose that image, audio or video content constituting a deep fake has been artificially generated or manipulated | For evidently artistic, satirical or fictional works, a disclosure that does not hamper the enjoyment of the work is enough. |
| Para. 4, second subparagraph | Deployer | disclose that text has been artificially generated where it is published to inform the public on matters of public interest | human review or editorial control, and a person holds editorial responsibility |
A deep fake under Article 3(60) of the AI Act is AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events. It would falsely appear to a person to be authentic. A generated product photo with no real-world counterpart does not meet this definition; a faked voice of the managing director does.
When does the AI labeling obligation apply?
The transparency obligations have applied since August 2, 2026, the general date of application under Article 113 of the AI Act. The Digital Omnibus Regulation (EU) 2026/1744 postponed the high-risk deadlines to December 2, 2027 and August 2, 2028. It did not postpone Article 50 (as of October 2026).
| Date | What applies |
|---|---|
| August 2, 2026 | Article 50(1) to (5) for all systems, existing ones included, unless a transitional rule applies |
| December 2, 2026 | End of the transition period under Article 111(4): providers of generative systems placed on the market before August 2, 2026 must implement machine-readable marking under paragraph 2 by this date |
The transition period covers only marking under paragraph 2 and only providers' existing systems. Deployers' disclosure obligations under paragraph 4 apply without a transition period. The postponed high-risk dates are explained in the article on the postponed AI Act deadlines.
Form: clear, distinguishable, at the first interaction at the latest
Article 50(5) of the AI Act requires the information to be clear and distinguishable and to be given at the first interaction or exposure at the latest. It must also meet the applicable accessibility requirements. The regulation does not prescribe a specific form. Examples of forms that meet these requirements:
- Chatbot: a visible notice before or with the first answer, such as "You are chatting with an AI assistant".
- Generated images and videos: machine-readable marking by the provider and, for a deep fake, a visible notice by the deployer.
- Voice and telephone systems: an announcement at the start of the call.
- Published texts on public matters: a notice on the text; where editorial control removes the duty, a documented review step with a responsible person evidences the exception.
The Commission supports a Code of Practice on transparency of AI-generated content and has created icons that deployers of generative AI systems may use. According to the Commission, adherence to the code is voluntary.
On July 8, 2026, the Commission concluded that the code adequately covers paragraphs 2, 4 and 5 (Commission assessment of the code). Signatories can rely on its measures to demonstrate compliance; adherence is not conclusive evidence of compliance.
Breaches fall under Article 99(4) of the AI Act
Article 99(4)(g) of the AI Act expressly lists breaches of the transparency obligations under Article 50. They therefore fall into the middle of the three fine tiers, below the prohibited practices under Article 5. For SMEs, Article 99(6) applies the lower of the two maximum amounts. Member States lay down the specific penalty rules under Article 99(1).
AI labeling and the GDPR apply side by side
Article 50(6) of the AI Act leaves other transparency obligations under Union or national law unaffected. If an AI system processes personal data, the information obligations under Articles 13 and 14 GDPR apply in addition. Examples are emotion recognition or a chatbot linked to a customer account.
Whether a data protection impact assessment is required is decided under the GDPR alone. The GDPR questions to settle when using AI assistants are covered in the article on ChatGPT and data protection.
Rizzqo carries role and output types for each AI system
Rizzqo carries on every AI system the company's role, such as provider or deployer, and the system's input and output types. Role and output type are the two variables that Article 50(2) and (4) turn on.
For this, the AI Act is modeled as a taxonomy on the asset model. Every chatbot, image generator and voice system appears as an asset with a named owner and an AI classification.