EU AI Act Article 50: Transparency Questions to Review

Start with scope and role

Article 2 includes providers placing AI systems or general-purpose AI models on the EU market, deployers located in the EU, and some providers or deployers outside the EU when the system's output is used in the Union. Applying those categories requires facts about the system, operator, market, location, and use of the output.

Article 50 distinguishes providers from deployers. A provider duty to design a system or mark its output should not automatically be assigned to every person who uses a third-party tool, and a deployer duty should not automatically be assigned to the model provider.

The Article 50 categories

Paragraph 1 addresses providers of systems intended to interact directly with people. Paragraph 2 addresses provider-side machine-readable marking for systems that generate synthetic audio, images, video, or text, subject to technical-feasibility language and stated exclusions.

Paragraph 3 addresses deployers of emotion-recognition and biometric-categorisation systems. Paragraph 4 addresses deployers of systems producing deepfakes and certain AI-generated or manipulated text published to inform the public on matters of public interest.

Timing, presentation, and exceptions

The Regulation generally applies from August 2, 2026. Article 50(5) says the specified information must be clear and distinguishable and provided no later than the first interaction or exposure, while also meeting applicable accessibility requirements.

Regulation (EU) 2026/1744 added Article 111(4): providers of systems covered by Article 50(2) that were placed on the market before August 2, 2026 have until December 2, 2026 to comply with that paragraph. This is a narrow transition for the provider-side machine-readable marking duty; it does not postpone Article 50(1), 50(3), or 50(4), and it does not cover a relevant system first placed on the market on or after August 2, 2026.

The text contains category-specific qualifications and exceptions, including language for artistic or fictional works and for certain public-interest text that undergoes human review or editorial control with identified editorial responsibility. Whether an exception fits cannot be decided from a single checker answer.

Regulation (EU) 2026/1744 also replaced Article 50(7). The current paragraph requires the Commission to encourage Union-level codes of practice, assess whether adherence is adequate to ensure compliance, and permits common implementation rules if a code is inadequate. This change is separate from the narrow Article 50(2) transition.

The European Commission's July 2026 guidelines explain how it interprets Article 50, but the guidelines are non-binding and do not change the Regulation. The Code of Practice remains voluntary. The Commission and AI Board formally assessed it as adequate for Article 50(2), (4), and (5), but the Commission says adherence is not conclusive evidence that a particular system complies; non-signatories may use other adequate means.

Questions for a professional review

Identify the system and output, each operator's role, where the operator is established, where the output is used, the purpose of publication, and whether the content is a deepfake or public-interest text. Record any human-review process and who holds editorial responsibility.

Also review other EU and national rules. Article 50(6) expressly leaves other transparency obligations unaffected, so satisfying one Article 50 step does not establish complete legal compliance. Article 50 violations fall within Article 99(4)'s maximum tier of €15 million or 3% of prior-year worldwide turnover for an undertaking, subject to the Regulation's case-specific, SME, and SMC rules; that ceiling is not a predicted penalty.

Frequently asked questions

Sources

• Regulation (EU) 2024/1689 — binding regulation
• Regulation (EU) 2026/1744 — binding regulation
• Consolidated Regulation (EU) 2024/1689 — documentation only non authentic
• Guidelines on Article 50 transparency obligations — nonbinding guidelines
• Code of Practice on Transparency of AI-Generated Content — voluntary code formally assessed adequate
• Commission opinion on adequacy of the Article 50 Code of Practice — formal adequacy assessment not conclusive compliance proof
• AI Board assessment of the Article 50 Code of Practice — formal adequacy assessment