EU AI Act Article 50: Transparency and Labelling Obligations Explained

Article 50 of the EU AI Act sets transparency obligations for AI systems that interact with people or generate content, and it applies from 2 August 2026. This guide summarises the European Commission’s Article 50 FAQ page (last updated 24 July 2026) and its page on the Code of Practice on Transparency of AI-generated Content. It is general information, not legal advice. For wider context on the Act, see our guide to the EU AI Act for finance professionals.

Who must do what

Under the Commission’s FAQ, obligations differ for providers and deployers.

• Providers of systems that interact directly with people (such as chatbots, agents and avatars) must design them so people are told they are dealing with AI, unless that is obvious. The FAQ says the disclosure must come at the start of the first interaction, be clear and distinguishable, and meet accessibility requirements, and that the "obvious" exception is interpreted restrictively. Systems working only in the background or machine-to-machine are out of scope.
• Providers of generative AI systems must ensure synthetic audio, image, video or text outputs are marked in a machine-readable format and detectable as AI-generated or manipulated. Exclusions listed include short strings of characters, source code, machine-to-machine outputs, closed-loop industrial outputs (unless final) and standard editing assistance.
• Deployers of emotion recognition or biometric categorisation systems must inform the people exposed to them that the system is operating.
• Deployers of deepfakes must disclose them using visible or audible labels at the latest on first exposure. Machine-readable marks added by providers do not satisfy this duty. For evidently artistic, creative, satirical or fictional works, disclosure must be appropriate and must not hamper the display or enjoyment of the work.
• Deployers publishing AI-generated or manipulated text to inform the public on matters of public interest must label it, unless it has had human review or editorial control and a person holds editorial responsibility. Spell-checking or grammar correction does not count as that review.

The Code of Practice

The Commission presents the Code of Practice on Transparency of AI-generated Content as a voluntary tool. Its page says the Commission and the AI Board confirmed it as an adequate voluntary tool for showing compliance, and it covers marking and detection by providers and labelling by deployers. The page reports about 190 organisations had signed by the end of July 2026. According to the FAQ, signatories can rely on the Code for legal certainty, while those who do not sign must show compliance by other adequate means and may face more information requests. For the duties on interacting systems and emotion recognition, the FAQ says providers and deployers choose their own measures. We have not read the text of the Code itself.

Timing and grace period

Article 50 applies from 2 August 2026. The FAQ says AI systems placed on the market before that date have until 2 December 2026 to comply, but only with the marking and detection duty in Article 50(2). Content generated before 2 August 2026 does not have to be labelled retroactively, although the Commission encourages it.

Penalties and enforcement

The FAQ says fines can reach €15 million or 3% of total worldwide turnover for the preceding financial year, with proportionality considered for small and mid-sized businesses. Enforcement is mainly by national market surveillance authorities. The AI Office has a limited role for certain systems built on general-purpose models and for systems integrated into designated very large platforms and search engines, and the European Data Protection Supervisor covers EU institutions. The FAQ does not say which authority is competent in each Member State.

Our suggestions

These are our own practical suggestions, not Commission guidance. List the AI tools your organisation provides to others and those you use to publish content. Work out whether any produce synthetic images, audio, video or public-interest text. For published text, decide who holds editorial responsibility and how review is recorded. Keep staff training current, as our article on AI literacy at work explains.

What the sources do not say

The FAQ page references Article 50(5) without describing it. It does not state whether the fine ceiling is the higher of the two figures or the tier that applies to Article 50 specifically, and it does not say whether the Article 50 guidelines are final; the Code page lists draft guidelines published on 8 May 2026 with a consultation. We have not read the Code or the guidelines in full.

Frequently asked questions

When does Article 50 apply?

From 2 August 2026, according to the Commission.

Do existing systems have extra time?

Systems placed on the market before 2 August 2026 have until 2 December 2026, but only for the marking and detection duty in Article 50(2).

Is the Code of Practice mandatory?

No. It is voluntary, although the Commission says signatories can rely on it for legal certainty.

Does AI-written text always need a label?

The FAQ says the duty covers text published to inform the public on matters of public interest, and it falls away where there is human review or editorial control with a person holding editorial responsibility.

Next steps

Explore our CPD resources to keep up with AI regulation, and check the Commission’s pages for current guidance before you rely on this summary.