Article 50
What are the AI transparency obligations under Article 50?
The obligations your customers see. Cheap to meet, highly visible to fail, and independent of risk classification.
Short answer
Article 50 of the EU AI Act imposes transparency duties regardless of risk tier. Providers must ensure people are informed when they are interacting with an AI system unless it is obvious, must mark synthetic audio, image, video and text content in a machine-readable format, and deployers must disclose the use of emotion recognition or biometric categorisation systems and must label deep fakes and, in some cases, AI-generated text published to inform the public on matters of public interest.
Reviewed 2026-08-02. General information for governance planning, not legal advice.
Key points
- Applies irrespective of whether a system is high-risk
- Interaction disclosure applies to customer-facing assistants and agents
- Synthetic media must be marked in machine-readable form
- Emotion recognition and biometric categorisation require disclosure to affected people
- Deep fakes require clear labelling
- Applies at full application, 2 August 2026
The four duties in practice
- Tell people when they are dealing with an AI system, not a person
- Mark synthetic or manipulated content so it can be detected downstream
- Disclose emotion recognition or biometric categorisation to those exposed
- Label deep fakes, and disclose AI-generated public-interest text where required
Why this is a brand risk, not only a legal one
Transparency failures are visible to customers, journalists and competitors in a way that documentation failures are not. An undisclosed AI agent handling complaints, or unlabelled synthetic imagery in a campaign, becomes a public story before it becomes a regulatory one.
The mitigation is unglamorous: know every customer-facing AI touchpoint, decide the disclosure wording once, and apply it consistently.
Building it into the register
Transparency should be a recorded attribute of each system, not a marketing decision made at launch. For each system the register should hold whether Article 50 applies, which limb applies, the disclosure implemented, where it appears and who owns it.
AIRAS Cloud captures transparency screening as part of assessment, so a system cannot pass review with an unresolved disclosure requirement.
Frequently asked questions
- Do we need to disclose a customer service chatbot?
- Yes, unless it is obvious to a reasonably well-informed person that they are interacting with an AI system. In practice a clear, persistent disclosure is the safer approach.
- Does Article 50 apply to internal-only tools?
- The interaction and content-marking duties are aimed at exposure to natural persons, but internal use involving emotion recognition or biometric categorisation carries its own disclosure duty, and workplace emotion inference is separately prohibited.
- What counts as machine-readable marking?
- Technical marking such as watermarking, metadata or provenance signals that allow synthetic content to be detected, applied in a way that is effective, interoperable and robust as far as technically feasible.
Primary sources
How AIRAS Cloud supports this
Related answers
Turn the regulation into an operating record
AIRAS Cloud gives Irish and EU organisations one accountable place to discover AI, determine scope, classify defensibly, assign controls and evidence every decision.
No pricing commitment. No confidential information required.