Many companies now use AI to write text and to create or edit images. In 2026, the EU made it a legal duty to tell people when they see some types of AI content. Denmark is adding its own rules about digital imitations of real people. This article gives a short overview. It is not legal advice.
Two roles: provider and deployer
The EU AI Act (Article 50) puts duties on two roles:
- Providers make the AI system, for example an image generator or a text model.
- Deployers use the AI system in their work, for example a company that publishes AI images in a campaign.
Most companies are deployers. Some companies are both.
Providers: machine-readable marking
Providers of systems that generate synthetic text, images, audio or video must mark the output in a machine-readable format. Software must be able to detect that the content is AI-generated or AI-manipulated. Typical methods are signed metadata and invisible watermarks.
The duty does not apply when the AI only does standard, assistive editing and does not substantially change the input or its meaning.
Deployers: visible labels
Deployers have two main duties:
- Deepfakes. If you use AI to create or change images, audio or video that look like real people, places or events and could appear authentic, you must disclose it. For evidently artistic, satirical or fictional work, a lighter disclosure is enough, as long as it does not spoil the work. Advertising and informative content do not get this lighter rule.
- Text on matters of public interest. If you publish AI-generated or AI-changed text to inform the public on matters of public interest, you must disclose it. The duty does not apply if a person has reviewed the text with real editorial control, and a person or company takes editorial responsibility for the publication.
The label must be clear and must come at the latest when the reader first sees or hears the content. If people can join a video or audio stream partway through, one notice at the start may not be enough.
Dates you need to know
| Date | What happens |
|---|---|
| 2 August 2026 | Article 50 applies. Deployers must label deepfakes and public-interest text. |
| 2 December 2026 | End of the grace period for machine-readable marking for generative systems already on the market before 2 August 2026 (from the Digital Omnibus). |
| 1 January 2027 | Planned start of the Danish deepfake rules, if the Folketing adopts the bill. |
| 2 February 2027 | Deadline for providers to support interoperable watermark detection. |
There is no retroactive labelling. For deepfakes, the date of generation counts. For public-interest text, the date of publication counts. Thus, text that you generated before 2 August 2026 and publish later must be labelled, unless the editorial exception applies.
Breaches of the transparency duties can lead to fines of up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher. For SMEs, the lower amount applies.
The Code of Practice
The European Commission published a Code of Practice on transparency of AI-generated content in June 2026, and found it adequate in July 2026. It has one section for providers (marking and detection) and one for deployers (labelling). The Code gives a common “AI” label for visible disclosure. Companies that follow the Code have a good basis to show compliance, but the Code does not guarantee it.
Denmark: protection against digital imitations
On 8 October 2026, the Danish Minister for Culture proposed changes to the Danish Copyright Act (bill L 49). The bill is still in the Folketing, and its content can change. As proposed:
- § 73 a: Do not make realistic, digitally generated imitations of a person’s personal, physical characteristics public without the person’s consent. The person can withdraw the consent.
- Exceptions: Caricature, satire, parody and criticism of power or society are permitted, unless the imitation is misinformation that can cause serious harm to the rights or essential interests of others.
- After death: The protection continues until 50 years after the year of death.
- § 65 a: A separate rule protects performers and artists against realistic digital imitations of their performances.
The Danish bill does not add a new labelling duty. It does not change the deepfake rule in the AI Act. A label does not replace consent: if you show a realistic imitation of a real person, you will need consent under the Danish rules and a label under the AI Act.
What we recommend
- Make a list of where your company uses AI to create or change text, images, audio and video.
- Find out if you are a provider, a deployer or both for each use.
- Label deepfakes clearly, also in advertising.
- Give each AI-assisted text that informs the public a named editor who reviews it, or label it.
- Get written consent before you publish a realistic imitation of a real person.
- Keep the metadata and watermarks from your AI tools. Do not remove them in your publishing pipeline.
Sources
- The EU AI Act’s transparency rules: a practical guide to Article 50
- Reed Smith: The Code of Practice adequacy decision and the final Commission guidelines on Article 50
- Article 50 marking after the AI Omnibus: what 2 December 2026 means
- Dansk Erhverv: Kulturministeren har fremsat forslag til ændring af ophavsretsloven
- Folketinget: L 49
Do you need help to find and label AI content in your workflows?
Contact us