When do EU AI Act Article 50 transparency obligations apply?

The core transparency obligations in Article 50(1), (3) and (4) of Regulation (EU) 2024/1689 — disclosing that a person is interacting with an AI system, labelling synthetic/deep-fake content, and notifying people subject to emotion-recognition or biometric-categorisation systems — apply from 2 August 2026 for both providers and deployers. These deployer obligations were not changed by the Digital Omnibus.

Does Article 50 apply to deployers or only to providers of AI systems?

Both. Article 50 places transparency duties on providers (e.g. marking AI-generated content) and on deployers (e.g. disclosing chatbot interaction, informing people about emotion-recognition, and labelling deep fakes they publish). A company using a third-party AI tool is typically a deployer and still has Article 50 duties.

What must be disclosed under Article 50?

Four things: (1) that a user is interacting with an AI system rather than a human, unless obvious; (2) machine-readable marking of AI-generated or manipulated audio, image, video or text (synthetic content); (3) notice to people exposed to emotion-recognition or biometric-categorisation systems; and (4) clear disclosure when AI-generated or manipulated content constitutes a deep fake, or text published to inform the public on matters of public interest.

Is there a separate deadline for provider watermarking of AI content?

Yes. The machine-readable marking obligation on providers of generative AI under Article 50(2) has a grace period for legacy systems already on the market until 2 December 2026, while the broader Article 50(1),(3),(4) obligations apply from 2 August 2026.

What are the fines for breaching Article 50 transparency obligations?

Article 50 sits under Article 99(4) of the EU AI Act: non-compliance with obligations other than the Article 5 prohibitions is subject to administrative fines of up to EUR 15,000,000 or 3% of total worldwide annual turnover, whichever is higher (for SMEs and start-ups, the lower amount applies).

Is the Article 50 Readiness Checker legal advice?

No. It is a free self-assessment tool that maps Article 50 to eight practical questions. It is general information, not legal advice, and answers stay in your browser unless you choose to email yourself the gap summary. Only courts and competent authorities make binding compliance determinations.

EU AI ACT — FREE TOOL

Article 50 Readiness Checker

Article 50 of Regulation (EU) 2024/1689 sets transparency duties for chatbots, emotion-recognition, biometric categorisation, deep fakes and AI-generated text/image/audio/video. It applies from 2 August 2026. Run the 8-question check below to see where you stand and which disclosure obligations still have gaps.

This is a self-assessment, not legal advice. It maps the four sub-paragraphs of Article 50 (interaction notice, machine-readable marking, emotion/biometric notice, and deep-fake or public-interest text disclosure) to eight practical questions a deployer can answer today. Your answers stay in your browser unless you choose to email yourself the gap summary.

Free tool

Article 50 Readiness Checker

Eight questions on your EU AI Act Article 50 transparency obligations (in application 2 August 2026). Answer honestly — you get a readiness score and the concrete gaps to close. Nothing is stored unless you enter your email.

1. Have you inventoried every AI system that interacts with people or generates content (chatbots, assistants, generative text/image/audio/video)?
Art 50 only bites on specific system types. You cannot disclose what you have not mapped.
2. Do users get a clear notice that they are interacting with an AI system when they hit a chatbot or AI assistant (Art 50(1))?
Required unless it is obvious from the context. The notice must be given at or before first interaction.
3. Is AI-generated or manipulated output marked in a machine-readable format (Art 50(2): C2PA, SynthID, watermark or embedded metadata)?
Providers of generative AI carry this duty; deployers must not strip the signal downstream.
4. Do you disclose when content is a deep fake or artificially generated/manipulated media (Art 50(4))?
Deployers of AI producing deep fakes must disclose the artificial origin to the audience.
5. For AI-written text published to inform the public on matters of public interest, do you label it as artificially generated (Art 50(4))?
Exception only where a human takes editorial review and responsibility for the text.
6. If you run emotion-recognition or biometric-categorisation systems, do you inform the exposed persons of their operation (Art 50(3))?
Plus lawful GDPR / Law Enforcement Directive processing of the personal data involved.
7. Have you collected provider/vendor attestations confirming how each third-party AI system meets its Article 50 marking duties?
You rely on providers for Art 50(2) marking; you need evidence they actually do it.
8. Do you keep a dated evidence pack (disclosure templates, screenshots, register) ready for auditors, procurement or investors?
Supervisors and buyers ask for demonstrable measures, not just intentions.

0/8 answered

Article 50 — key facts

Article 50(1),(3),(4) transparency obligations apply from 2 August 2026 for both providers and deployers.
Provider machine-readable marking of synthetic content (Article 50(2)) has a legacy grace period until 2 December 2026.
These deployer transparency obligations were not changed by the Digital Omnibus.
Breaches fall under Article 99(4): up to EUR 15,000,000 or 3% of worldwide annual turnover (whichever is higher; lower for SMEs).

Want the detail behind each question?

Read the full deployer breakdown in the Article 50 transparency guide, and see the penalty exposure for getting transparency wrong in the EU AI Act fines guide.