Denmark's likeness right covers everyone
Denmark's copyright amendment covers every person's face, voice and body, not just public figures. Why the recognisability test creators rely on stops working.
Start with the part that nearly every write-up of Denmark's deepfake amendment gets wrong: the date. Depending on which source you read, the amendment to the Danish Copyright Act entered into force on 31 March 2026, or on 1 July 2026, or is still not in force at all. Those are not the same claim, and I could not resolve them against a primary source. So before anything else, here is the honest split between what is verifiable and what isn't — because the substance is worth planning around regardless of which date turns out to be right, and the substance is the part almost nobody reads.
What's verified, and what isn't
Verified. The European Parliamentary Research Service briefing on the Danish approach (PE 782.611, published January 2026) describes amendments to the Danish Copyright Act that were notified to the European Commission. They come in two parts: general protection against realistic digitally generated imitations of personal characteristics, and a separate protection against realistic digitally generated imitations of performers. Both require consent from the person imitated before the content is made available. Protection runs for 50 years after that person's death. Expressions of caricature, satire, parody and pastiche are in principle excluded.
Not verified. Entry into force. The EPRS briefing, writing in January 2026, said the amendments were "expected to enter into force in July 2026." Danish legal commentary published later described the amendment as still not in force. Other secondary sources give 31 March 2026. As of 20 August 2026 I could not find a source that confirms the amendment is operative today, and I am not going to assert one. If you need the answer for a live decision, check the Folketing record rather than any summary, including this one.
The reason to write about it anyway: this design is being copied. The EPRS briefing notes the Danish Minister of Culture stated that Ireland intends to follow the planned Danish rules, and that a declaration agreed under the Danish Council presidency — carrying the principle that citizens must be protected against digital replicas of their personal characteristics without consent — was signed by ministers from every Member State except Hungary, plus five other European countries. The specific commencement date matters much less than the fact that this shape of right is where Europe is heading.
Why the public-figure test stops working
Most creator-side risk assessment for synthetic media runs an informal three-part test, and Denmark's design breaks all three parts at once.
Test one: "is this person famous?" The working assumption behind a lot of production practice is that likeness risk scales with recognisability — that a celebrity has a claim and an anonymous face doesn't. This is roughly how right-of-publicity thinking has trained people to reason, because publicity rights historically attached to commercial value in an identity, and commercial value clusters at the famous end. Denmark's protection is written for all natural persons. There is no fame threshold to fall below. An entirely ordinary-looking generated face that happens to be a realistic imitation of a specific real person is inside the protection on exactly the same terms as a household name.
Test two: "is it harmful?" The second filter is that flattering or neutral output is safe, and only defamatory or degrading output creates exposure. Legal commentary on the amendment — including the Verfassungsblog analysis of Denmark's approach — describes compensation being available without proof of reputational harm, alongside takedown and platform liability. If that reading holds, a pleasant, complimentary, entirely well-intentioned synthetic clip of someone who didn't consent sits in the same place as a nasty one. The consent is the operative fact, not the sentiment.
Test three: "is it obviously not real?" Obviousness is not the carve-out. Caricature, satire, parody and pastiche are excluded in principle, but those are specific, judged categories with a long body of interpretation behind them — not a general defence that anything stylised or clearly synthetic falls outside. "Nobody would think this was really them" is not the same argument as "this is parody," and only the second one is in the text.
Notice what survives all three. The only reliable filter left is documented consent from the specific person being imitated, which is precisely the thing most fast creative pipelines don't have and can't produce retrospectively.
Two protections, not one
The performer provision is easy to skim past and it's the one that catches commercial work. The general protection covers personal characteristics — appearance, voice, movement. The performer protection covers realistic digital imitations of a performer's performance, which is a different object: not just what someone looks like, but the specific delivery, phrasing and physical style that makes their work theirs.
That distinction matters if you generate voice. A voice model trained toward a recognisable delivery, or a prompt that reaches for one, engages a performer's interest even when the resulting audio isn't a technical clone of any single recording. Voice likeness law for creator marketing covers the broader shape of this, and it's the axis with the least tooling built around it — face-covered, voice-exposed is the gap in plain terms. Most likeness-detection infrastructure watches for faces. Almost none of it listens.
Enforcement runs through the DSA
The Danish approach isn't primarily a plan to litigate in Danish courts. Per the EPRS briefing, enforcement is intended to run through the EU Digital Services Act, which already sets Union-wide rules on detecting, flagging and removing illegal content — including copyright-infringing content, defined at either EU or national level. The Danish draft is designed to give a legal basis for activating those notice-and-action procedures against non-authorised imitations.
That is a meaningfully different threat model from a lawsuit. It means the practical consequence of getting this wrong is a takedown request that a platform is obliged to process, arriving through machinery that already exists and already works, against a live campaign. Speed, not damages, is the exposure.
The same briefing is candid that the framing is contested. It notes that copyright typically protects original works reflecting an author's own intellectual creation, that voice and image considered in isolation don't naturally fall into that category, and that the Danish authorities' own explanatory notes state they do not intend to create a new right of copyright for citizens — the amendments sit inside the Copyright Act but function as a supplement to private-law personality rights. Several Member States, per the same document, argued the problem extends beyond copyright's remit entirely.
What to change in a real workflow
Three things, and none of them are new legal process:
1. Move consent upstream of generation, not downstream of it. If a real person's face, voice or performance is anywhere in the input — a reference image, a voice sample, a scanned likeness — the consent record needs to exist before the generation runs, not before publication. Retroactive consent is the thing that isn't available. Usage rights in creator contracts is the practical version, and usage rights is the concept in one page.
2. Treat "not a public figure" as no longer being a green light. Anywhere your process has a step that effectively asks "is this person famous enough to matter," that step is now checking the wrong thing. The question is whether the output is a realistic imitation of an identifiable person, full stop.
3. Separate synthetic-from-scratch from synthetic-of-someone. A generated presenter who is not an imitation of any specific real person sits in a completely different position from a digital replica of a real one. If you use AI avatars or voice cloning, knowing which of those two things each asset is — and being able to show it — is the whole audit. Enrolling in likeness detection is worth doing for the people whose likeness you legitimately license, so you find out fast when someone else uses it.
FAQ
Is Denmark's likeness amendment in force right now?
I could not verify that. Published dates conflict across 31 March 2026, July 2026 and "not yet in force," and no source I found confirms entry into force as of 20 August 2026. Check the Folketing record before relying on the answer either way.
Does it only protect Danish citizens?
The protection as described applies to natural persons, and legal commentary discusses it reaching likenesses reproduced or disseminated within Danish jurisdiction. Territorial reach for content originating outside Denmark is one of the practical enforcement questions the EPRS briefing flags as unresolved.
Does a parody exception cover comedic AI content?
Caricature, satire, parody and pastiche are excluded in principle. That's a defined set of categories, not a general exemption for anything humorous or obviously synthetic. Assuming the exception applies because output looks unreal is not the same as the exception applying.
Does this affect a fully invented AI presenter?
Not on its face. The protections attach to realistic imitations of identifiable people. A synthetic presenter that isn't an imitation of any specific real person isn't what the amendment is aimed at — which makes the distinction between "generated" and "generated from someone" the single most useful thing to record about every asset you ship.
The date will resolve itself. The design won't change: consent from a named individual, obtained before generation, is the only test that passes all three of the filters this framework removes.