Deepfake laws across Europe: what do they look like and what do they mean for Wikipedia

Deepfakes are understood as realistic images, audio and video of real people that are generated or manipulated with AI. They can be used for parody, satire or to express valid social criticism. But they can be harmful in the cases of fraud, economic damages, disinformation and non-consensual intimate imagery.

Lawmakers in several EU countries (Belgium, Denmark and Cyprus) are trying to answer with new rights over a person’s likeness and voice, usually placed in copyright or neighbouring-rights law.

For the Wikimedia projects, which host and reuse media under free licences and also document public events and debates (including deepfake incidents themselves), the details of these laws matter. So let’s take a closer look!

Belgium: a new neighbouring right over image and voice

In September 2026 the Belgian Federal Chamber published a bill amending the Code of Economic Law with regard to deepfakes. It would insert a new Article creating a neighbouring right.

What it does

Every natural person gets an inalienable moral right over their image and voice, plus an exclusive right to authorise or prohibit the making, reproduction, distribution and communication to the public of a deepfake that uses them.
A deepfake is defined as AI-generated or AI-manipulated image, video or audio that resembles an existing or deceased person so closely that it may wrongly be taken for authentic.
The rights last 50 years after death and are then exercised by heirs. They can be licensed against appropriate remuneration, and collective management organisations can manage them.
Courts can order urgent provisional measures, including seizure and the blocking of accounts or domains, and can order removal and de-listing from search results. Online intermediaries must carry out such court orders without delay, “without prejudice” to the Digital Services Act (DSA).

The exceptions. There is no infringement where the person has consented, or where the deepfake is used for parody, satire or social criticism, provided a fair balance is struck with freedom of expression and the use does not disproportionately harm the person’s honour or reputation.

Our first reading: some reassurances, but incomplete. The parody and satire exceptions are welcome, and the court-ordered enforcement model is narrower than a duty to filter. But several points could affect Wikimedia projects:

A closed list of purposes. There is no explicit exception for news reporting, quotation, education, research, documentation or archiving. An encyclopaedic article about a notorious deepfake may need to show or link to it. That is a case where neither parody nor social criticism fits.
A balancing test. Even where an exception applies, it depends on a judge finding a “fair balance”. This creates a a lot of uncertainty, including for good-faith actors.
A broad definition. The definition requires neither intent to deceive, nor harm, nor absence of a label. “Manipulated by AI” could, depending on interpretation, reach AI-assisted restoration, colourisation or upscaling of historical photographs and footage. Protection also extends to deceased persons for 50 years.
Consent versus free licensing. Files on Wikimedia Commons are reused far beyond the Wikimedia projects. Will this new neighbouring right work with general public lincences?

Denmark: copyright as the vehicle

The Danish Ministry of Culture’s bill amends the Copyright Act in two ways. A new section protects any person against the making available of realistic digitally generated imitations of their personal characteristics (appearance, voice) without consent. Another new section gives performing artists a similar protection against imitations of their performances. Protection would run for the person’s lifetime plus 50 years and applies in Denmark only. The exceptions cover caricature, satire, parody, pastiche, criticism of power, social criticism and similar uses.

The draft went to public consultation from 7 July to 21 August 2025 and was then notified to the Commission under the TRIS procedure (notification 2025/0654/DK, received 31 October 2025, a re-notification of a draft that had been withdrawn earlier). The planned entry into force was 31 March 2026. As far as we can tell, the bill has not been adopted and is stuck in the process.

The copyright route is deliberate: it lets victims use the notice-and-action mechanism of the DSA, and it would be tied into the Danish rule implementing Article 17 of the Copyright in the Digital Single Market (DSM) Directive.

Our reading: better drafted than the Belgian text, but not yet completely safe: The list of exceptions is broader (it includes pastiche and criticism of power), but our Danish chapter Wikimedia Denmark’s own conclusion was that the bill in its current form would limit the possibilities. The main concerns:

Informative, documentary and artistic use is not covered. WMDK points out that Commons holds realistic AI imitations of known people: artworks, satire or criticism of power, and “neutral” images used to illustrate articles on deepfakes and generative AI. Only the satire and criticism group clearly fits the exceptions, hence WMDK’s request to add imitations with a predominantly informative, educational or documentary purpose.
The exceptions turn on judgement calls. The exception is framed around imitations that are predominantly an expression of the listed purposes, with a further test of concrete risk of serious harm. In its own consultation response, the media association Danske Medier said these formulations require assessments that will be very difficult to make in practice.
Filtering pressure. The European Commission’s TRIS comments point to the risk of extended filtering duties.
Fragmentation. Commons projects are normally global in nature. How these special, national rules sit with a single, global repository is an open question.

Cyprus: A new sui generis right already law

In March 2026 the Cypriot House of Representatives adopted law amending its copyright rules. It creates a new sui generis right against imitations, through “deepfake products” (a term that cross-refers to Article 3(60) of the AI Act), of a person’s physical characteristics or of a performer’s performance. According to Tatiana Synodinou’s analysis on the Kluwer Copyright Blog, the text follows the Danish bill closely but is less elaborated, and it was enacted without public consultation or parliamentary debate.

Scope. The law prohibits making a deepfake imitation of a performer’s performance available to the public without express consent. To achieve this it gives every natural person a protection for their physical characteristics (morphology, voice, biometric and behavioural features), which lasts for 50 years after death.
Enforcement. The rules are those of copyright enforcement: civil action, damages and injunctions, plus criminal sanctions.
Exceptions. There is an exemption for satire and parody, caricature and political or social criticism. Both are overridden where the imitation amounts to “disinformation” that creates a “serious risk” to the rights or essential interests of others. No exception for informative, educational or documentary use.
Licensing. The amendments don’t say anything on whether the right can be licensed or transferred.

Our reading: problematic in several ways:
For Wikimedia, lack of exceptions for informative, educational or documentary use are problematic. Additionally, the “disinformation” and “serious risk” override makes even existing exceptions hard to predict. Another worry is that national sui generis rights can lead to fragmentation of law and the internet.

Do we need another layer?

Let’s take the most recent example. The Belgian memorandum argues that the existing framework is inadequate: the AI Act requires labelling but does not ban deepfakes, the GDPR protects only to a degree, and Belgian portrait and performers’ rights do not fit. These are fair observations. But the question we would put to every legislator is whether the gap is legal or practical.

The current toolbox is already substantial and growing:

AI Act transparency rules on labelling deepfakes, and the agreed ban on AI “nudify” systems under the Digital Omnibus on AI;
DSA notice-and-action and risk-mitigation duties for platforms;
the GDPR, including erasure rights for personal data such as image and voice;
national personality, image and defamation laws, and criminal law on fraud and non-consensual intimate imagery, with EU rules requiring Member States to criminalise sexual deepfakes.

The worst harms (scams, sexual abuse imagery, election manipulation) are not primarily about someone’s lack of control over a commercial market in their likeness. Exclusive, licensable and inheritable rights, as in Belgium and Cyprus, create such a market and then require every legitimate use to fit an exception. A patchwork of 27 national regimes also sits poorly with a borderless internet and with the AI Act’s harmonised approach.

We are not convinced that a new layer of protection is the right instrument here. Filling the potential gaps of currently existing instruments might be much more efficient. Whatever we do, we should at minimum:

include open-ended public-interest exceptions for news, education, research, documentation and encyclopaedic use, quotation and archives, not only parody and satire;
tie liability to deceptive or harmful use;
keep intermediary obligations within the DSA liability regime;