EU AI Act transparency and AI-content labelling: what applies by Dec 2026

What Article 50 requires

This guide uses four European Commission pages: the Code of Practice page (last updated 19 October 2026), the guidelines page (6 August 2026), the Quick Facts page (29 July 2026) and the EU icons page (13 October 2026). Where the guide says "the Commission says", it is quoting one of these pages.

The Commission's guidelines page says Article 50 of the AI Act applies from 2 August 2026. The Quick Facts page sets out four cases. Two fall on providers of AI systems and two on deployers.

For providers, the guidelines page says they must design AI systems so that individuals are told when they interact with an AI system directly. The second provider duty is marking AI-generated content, covered in the next section.

For deployers, the guidelines page lists three situations where people must be informed: emotion recognition and biometric categorisation tools, deepfakes, and text publications on matters of public interest without human review or editorial control.

Provider duties: marking and detection

The guidelines page says providers must add machine-readable marks so that AI-generated or manipulated content can be detected. The Quick Facts page names the content types as text, image, video and audio.

The Quick Facts page also states an exception. A provider is not required to mark where the system performs an assistive function for standard editing or does not substantially alter the input data.

The Code of Practice page describes what the provider working group focused on. Its target is that outputs in audio, image, video and text are marked in a machine-readable format and can be detected as artificially generated or manipulated.

Deployer duties: deepfakes and public-interest text

Two kinds of content trigger the deployer disclosure duty under Article 50(4). The first is deepfakes. The EU icons page defines a deepfake as AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear authentic or truthful.

The second is text. The EU icons page describes it as AI-generated or manipulated text published to inform the public on matters of public interest. The duty applies where the text had no human review or editorial control and no person assumed editorial responsibility.

The EU icons page lists three limits on these duties:

• For evidently artistic, creative, satirical or fictional works that contain deepfakes, the page says the obligation is limited to disclosure that does not hamper display or enjoyment of the work.
• The page says the duty does not apply where law authorises the use for criminal law purposes.
• For text, the page says the duty does not apply once the text has had human review or editorial control and a person holds editorial responsibility.

The guidelines page says the guidelines give examples of what is in and out of scope, including standard editing as an exception. Teams deciding whether a given output is covered are directed to the guidelines for those examples.

The Code of Practice and how to show compliance

The Code of Practice on Transparency of AI-generated Content is split into two sections. Section 1 covers providers and the marking and detection of AI-generated content. Section 2 covers deployers and the labelling of deepfakes and AI-generated text. The Commission's Quick Facts page says the code was published in June 2026.

Signing is optional, but the legal duties are not. The Commission says so directly.

The Commission says it and the AI Board have assessed the code positively. The Commission says signatories can rely on the code's measures to demonstrate compliance with the marking and labelling rules.

A provider or deployer that does not sign faces a different test. The guidelines page says they must show compliance with the marking and labelling duties by alternative means that are equivalently adequate. The Code of Practice page adds that market surveillance authorities assess such measures individually.

For the other transparency duties, such as telling people they are interacting with an AI system, the guidelines page says providers and deployers choose adequate measures themselves, taking the guidelines into account.

The Code of Practice page also says the code does not replace the AI Act or the Commission's guidelines on Article 50.

The December 2026 marking deadline

The Quick Facts page sets out one grace period. It covers the marking obligation only, and only for generative AI systems placed on the market before 2 August 2026. The page gives the end of the period as "December 2026" and cites Article 50(2) as amended by the AI Omnibus.

None of the four pages gives a day within December 2026. The guide therefore cannot confirm 2 December from these sources.

The Quick Facts page also addresses older deepfakes. Labelling is not mandatory for deepfakes generated before 2 August 2026, but it is encouraged.

The documents describe the grace period only for the marking obligation, which sits with providers. They do not describe a grace period for deployer labelling. They also do not give a checklist for teams. What they do set out is the duties above and the two routes to show compliance: signing the code, or demonstrating equivalent measures.

Labels and the EU icons

The EU has published icons that deployers may use to label content. The Commission says use of the icons is optional but the labelling duty is not. The Quick Facts page says there are three icons. The EU icons page names them as the Basic icon, Fully AI-Generated and Partially AI-Modified, each offered in black, white, and 50% transparent versions.

The icons page says the icons do not by themselves establish compliance, and deployers remain responsible for meeting Article 50. It gives a summary, which it says is not a complete overview, of display rules. Signatories are directed to Section 2 of the code for the detailed placement specifications.

• The icon should be clearly perceivable and distinguishable at the latest at first exposure.
• The icon must be visible when content is reshared or downloaded.

The icons page also reports user testing. It says performance improved on all measures when the basic icon came with a text label.

Enforcement and penalties

The Quick Facts page names three enforcers: national market surveillance authorities, the AI Office for systems under its supervision, and the European Data Protection Supervisor when EU institutions are providers or deployers.

On penalties, the page lists fines of up to €15M, or up to 3% of total worldwide annual turnover for companies. For EU institutions, bodies and agencies the page lists fines of up to €750k. The page also notes proportionality for SMEs and SMCs.