In force since 2 August 2026

Article 50: the AI Act's transparency duties

If your AI talks to people or produces content they see, they generally need to be told. Here's what that means in practice and who is responsible.

Last checked 3 October 2026

The four duties

SituationDutyFalls on
AI that interacts directly with people, such as a chatbot or voice agentPeople must be informed they're dealing with AI, unless it's obvious from the contextProvider Art. 50(1)
AI that generates images, audio, video or textOutput must be marked in a machine-readable way as AI-generatedProvider Art. 50(2)
Emotion recognition or biometric categorisationThe people exposed must be informedDeployer Art. 50(3)
Deepfakes, and AI-written text published to inform the public on matters of public interestDisclose that the content is AI-generated or manipulatedDeployer Art. 50(4)

Information must be given clearly, at the latest at the first interaction or exposure, and in an accessible way.

"Provider" or "deployer": why it matters

If you bought a chatbot off the shelf, the vendor is usually the provider and must build the disclosure in. You're the deployer, and your job is to make sure it's switched on and works on your site.

If you built your own assistant on top of a model, or put a tool on the market under your own name, you may be the provider yourself, and the duties in 50(1) and 50(2) are yours.

Getting the role right is the part most businesses skip. It decides whose duty it is, and who answers for it if a regulator asks.

Common SME situations

A chatbot on your website

Customers need to know they're talking to AI. A clear line at the start of the conversation usually does it. Check that your vendor's widget shows one and that you haven't hidden or reworded it.

AI-written marketing content

Ordinary marketing copy written with AI help isn't automatically caught. The specific duty for text covers content published to inform the public on matters of public interest, and doesn't apply where a person has reviewed it and takes editorial responsibility. AI-generated or altered images, audio or video that look real are a different matter: those count as deepfakes and must be disclosed.

AI voice agents on the phone

Callers must be told they're speaking to an AI unless it's obvious. Put the disclosure in the greeting.

Grace period for existing generative AI

Under the Digital Omnibus, generative AI systems placed on the market before 2 August 2026 have until 2 December 2026 to meet the machine-readable marking duty in Article 50(2). The other transparency duties have applied since 2 August 2026.

Penalties

Breaching Article 50 can lead to fines of up to €15 million or 3% of worldwide turnover. For SMEs, the cap is whichever is lower (Art. 99(4) and 99(6)).

Not sure which of your tools are covered? Take the two-minute check or book a call.

Sources: Regulation (EU) 2024/1689, Article 50 and Article 99, as amended by the Digital Omnibus on AI. The Commission is preparing guidelines and a code of practice on Article 50, and we'll update this page when they're final. General information, not legal advice.