Is your AI meeting Article 50 — in force since 2 August 2026?
The EU AI Act's transparency obligations have applied since 2 August 2026 — chatbot disclosure, machine-readable marking of AI-generated content, deepfake labels and emotion-recognition notices. They are enforceable today, so anything unmet is a live exposure rather than a deadline to plan for. Answer seven yes/no questions to get your tailored duty list and a week-one action plan. Deterministic, no sign-up.
1.Do you operate an AI system that interacts directly with people in the EU (chatbot, voice assistant, AI agent)?
Article 50(1) — users must be informed they are interacting with AI, unless it is obvious from context.
2.Does your system generate synthetic audio, image, video or text content?
Article 50(2) — providers of generative systems must mark outputs as artificially generated.
3.Are those outputs marked in a machine-readable format as AI-generated?
Only relevant if the question above is answered yes.
4.Do you publish deepfakes — AI-generated or manipulated content resembling real people, places or events?
Article 50(4) — deepfakes carry a visible disclosure duty on top of machine-readable marking.
5.Are those deepfakes visibly labeled as artificially generated or manipulated?
Only relevant if the question above is answered yes.
6.Do you use emotion recognition or biometric categorisation systems?
Article 50(3) — subjects must be informed. Workplace and education uses of emotion recognition are prohibited outright (Article 5).
7.Do you inform the people exposed to those systems about their operation?
Only relevant if the question above is answered yes.
Answer the questions above to see your Article 50 duty list. Nothing is stored or sent anywhere.
What Article 50 covers
Article 50 is the EU AI Act's transparency layer — the tier that catches most everyday AI products that are neither prohibited nor high-risk. Since 2 August 2026 it has required:
- • Art. 50(1) — AI systems interacting with people must disclose they are AI.
- • Art. 50(2) — providers of systems generating synthetic audio, image, video or text must mark outputs in a machine-readable format (transition to 2 December 2026 for systems already on the market).
- • Art. 50(4) — deepfakes must be visibly labeled as artificially generated.
- • Art. 50(3) — people exposed to emotion recognition or biometric categorisation must be informed.
Breaches sit in the €15M / 3% penalty tier — and the same date activated the Commission's GPAI enforcement powers under Article 101. Read what actually changes on 2 August 2026 and the content-marking guide, or run the full risk checker to classify your system across all tiers.
Article 50 questions, answered
When does Article 50 of the EU AI Act apply?+
Article 50 transparency obligations apply from 2 August 2026. There is one transition inside it: providers whose generative systems were already placed on the market have until 2 December 2026 to implement machine-readable marking under Article 50(2). Everything else — chatbot disclosure, deepfake labels, emotion-recognition notices — applies in full from 2 August 2026.
Did the 2026 Digital Omnibus delay Article 50?+
No. The Digital Omnibus moved the high-risk dates — Annex III obligations to 2 December 2027 and high-risk safety components of regulated products to 2 August 2028 — but Article 50 transparency was not postponed and still applies from 2 August 2026, as do the Commission's GPAI enforcement powers.
What does Article 50 actually require?+
Four duties. Article 50(1): AI systems that interact directly with people must be designed so those people are informed they are dealing with AI. Article 50(2): providers of systems generating synthetic audio, image, video or text must mark the outputs machine-readably. Article 50(3): people exposed to emotion recognition or biometric categorisation must be informed. Article 50(4): deepfakes must be visibly disclosed as artificially generated or manipulated.
Does my chatbot need an AI disclosure?+
If it interacts directly with people and is used in the EU, yes — Article 50(1) requires that users be informed they are interacting with an AI system. The only carve-out is where that is obvious to a reasonably well-informed, observant and circumspect person. A conversational assistant that writes like a colleague is not obvious in that sense, so the safe course is an explicit notice at the start of the interaction.
What counts as a deepfake under Article 50(4)?+
AI-generated or AI-manipulated image, audio or video content that resembles real persons, objects, places, entities or events and would falsely appear authentic. The deployer must visibly disclose that it is artificially generated or manipulated. Evidently artistic, creative, satirical or fictional work may present the disclosure in a way that does not spoil the work, but the disclosure is still required.
What is the 2 December 2026 date about?+
It is the transition for machine-readable content marking under Article 50(2), and it only helps generative systems that were already on the market before 2 August 2026. Systems placed on the market after that date have to mark outputs from launch, and the transition does not defer chatbot disclosure, deepfake labels or emotion-recognition notices at all.
Is Article 50 the provider's job or the deployer's?+
Both, split by paragraph. Article 50(1) and 50(2) sit with providers — designing the interaction disclosure into the system and marking generative outputs. Article 50(3) and 50(4) sit with deployers — informing the people exposed to emotion recognition or biometric categorisation, and visibly labelling deepfakes they publish. Many companies are both at once for different systems.
What is the penalty for breaching Article 50?+
Article 50 breaches fall in the middle penalty tier of Article 99: up to €15 million or 3% of worldwide annual turnover, whichever is higher — and for SMEs, whichever is lower. That is the same tier as most other operator obligations; only the Article 5 prohibitions carry the €35M / 7% maximum.