Home › Guides

EU AI Act Article 50: disclosure obligations for AI-dubbed feature distribution in 2026

By the Fonti Studio post-production team · 11 June 2026

An AI dub is, in the language of the regulation, a deep fake. The synthetic voice resembles a real person, the original actor, and a listener would take it to be authentic. That single classification is what pulls a French AI dub of an English feature inside Article 50 of Regulation (EU) 2024/1689, and it is why, from this August, disclosure stops being a courtesy line in the press notes and becomes part of the deliverable.

This piece is for the distributor or sales agent who licenses an AI-dubbed master into an EU territory and has to answer a buyer's compliance team. Two questions matter. Who has to put the AI mark on the file, and who has to tell the viewer. They are not the same party, and the obligations do not transfer with the license.

The 2 August 2026 trigger

Article 50 carries the transparency obligations of the AI Act, and it applies from 2 August 2026, twenty-four months after the Regulation entered into force. The European Commission's article reference gives the date plainly; the AI Act Service Desk run by the Commission carries the operative text.

Nothing about the underlying obligation is new on that date. What is new is enforceability. Up to 2 August the disclosure of a synthetic voice is good practice and a platform preference. After it, it is law across the member states, with national market-surveillance authorities behind it. A distributor who ships an undisclosed AI dub into Belgium, France, or Germany on 3 August is non-compliant in a way that did not exist on 1 August. That is the overnight change.

Two parties, two obligations

Article 50 splits the duty in a way that maps almost too neatly onto a dub-distribution chain.

The provider of the AI system, the party that generates the synthetic content, sits under Article 50(2): the output has to be "marked in a machine-readable format and detectable as artificially generated or manipulated," and the marking has to be "effective, interoperable, robust and reliable as far as this is technically feasible." This is a file-level obligation. It lives in the bytes, not on the screen.

The deployer, the party that puts the content in front of an audience, sits under Article 50(4): a deployer "that generates or manipulates image, audio or video content constituting a deep fake, shall disclose that the content has been artificially generated or manipulated." This is a viewer-facing obligation. It lives on the screen, or in the credits, or on the landing card.

In a dub workflow the production house that renders the synthetic voice is closest to the provider role, and the distributor who releases the dubbed master into a territory is closest to the deployer. The Commission's own framing, in the draft guidelines published 8 May 2026, keeps the two distinct: providers design and mark, deployers disclose. The deployer's compliance is not contingent on the provider having done its job, and it does not transfer up or down the chain by contract. A buyer cannot license away the obligation to tell the viewer.

So the distributor needs two things in hand before release: a file that is already machine-marked, and a disclosure surface they control in their own territory.

The mark we put on every Fonti Studio dub

We treat the provider-side obligation as ours to discharge, because we are the party generating the synthetic voice. Every Fonti Studio dub deliverable ships marked.

Concretely, that is two layers. An imperceptible watermark in the dubbed audio stem, which is the machine-readable, file-level signal Article 50(2) asks for. And a header comment in every subtitle source file that records the AI use, the pipeline, and the date, so the disclosure travels with the text track as plain provenance metadata. The audio watermark is the load-bearing one for 50(2); the source-file header is belt and braces for any QC operator who opens the file.

The Regulation is technology-neutral and does not name a watermarking standard in its operative text. It describes the property it wants: machine-readable, effective, interoperable, robust, reliable. The Commission's Code of Practice on transparency of AI-generated content, released by the AI Office on 10 June 2026 with an initial-signatory deadline of 22 July 2026, points implementers toward C2PA Content Credentials as the favoured carrier. We track that Code as the marking spec moves from "favoured" to "expected," and our provider-side disclosure is written into Terms §4.4 so a buyer can see the commitment before they sign.

This is the same mark that rode along with the Kumite reference dub, where the cloned French track measured 0.23 speaker similarity to the original cast against 0.05 for studio-preset voices. The mark does not degrade the dub. It is metadata, not noise.

The disclosure language we draft for distributors

The viewer-facing half is the distributor's, and here the AI Act is gentler on film than the headlines suggest.

A feature is, in the words of Article 50(4), an "evidently artistic, creative, satirical, fictional or analogous work." For that category the deep-fake disclosure obligation is explicitly "limited to disclosure of the existence of such generated or manipulated content in an appropriate manner that does not hamper the display or enjoyment of the work." The Commission's draft guidelines illustrate the point with an unobtrusive label in the credits or on a loading screen, rather than an immersion-breaking marker burned into the frame.

That is the difference between a watermark stamped over the actor's face for ninety minutes and a single clean line in the end credits. The artistic-work carve-out means a dubbed feature does not need an in-frame "AI" badge. It needs the existence of the AI dub stated where a viewer can find it without the film stopping to tell them.

We draft that line on request. A credit-roll card and a one-line platform-metadata field, sized to the territory and the release format, in the local language. It is short by design. The Regulation does not reward over-disclosure, and a wall of legal text in the credits hampers enjoyment, which is the one thing the carve-out tells you not to do.

Where Article 50 meets the platforms

Platform policy moved ahead of the law, which simplifies things. Netflix published its guidelines for the use of generative AI in content production in August 2025, asking partners to flag any GenAI use to their Netflix contact and reserving written approval for final deliverables that involve talent likeness. A distributor who already routes their GenAI use through that partner channel is most of the way to an Article 50 posture, because the platform conversation surfaces the dub before the file is ever delivered.

The interaction to watch is that platform policy and the AI Act are not the same instrument. A platform can accept an AI dub on its own terms and the distributor still owes the territory-level consumer disclosure that Article 50(4) imposes, because the Regulation binds the deployer directly. Clearing the platform's GenAI review is necessary. It is not sufficient. The legal surface is the viewer in the member state, not the acceptance email from the platform's delivery team.

What we do not warrant

We mark the file. We do not, and cannot, warrant the distributor's territory-specific consumer surface, because we do not control it. The end-credit card in a Belgian theatrical run, the metadata field in a French VOD storefront, the on-screen notice a national authority may later specify in guidance: those are authored by the party releasing into that territory. The deployer obligation is theirs by the structure of the Regulation, not by anything we could sign on their behalf.

What we hand over is everything that makes their disclosure easy to make: a marked deliverable, documented provenance, and draft language to drop into the release. The judgement call about exactly where the line lands in each territory stays with the distributor, who knows their windows and their local counsel.

On the release deck, day one

The practical answer to "what goes on the deck the day you license an AI-dubbed feature into the EU" is three lines.

One, the master is machine-marked under Article 50(2): watermark in the audio, provenance in the source files. Two, the consumer disclosure is the distributor's to surface, in the attenuated artistic-work form, a credit-roll line plus a metadata field, not an in-frame badge, and it is drafted and ready in the territory language. Three, the platform's own GenAI process has been cleared in parallel and does not replace line two.

A Belgian theatrical release is the clean worked example. The dub is delivered marked. The disclosure rides in the end credits as a single localised card and in the cinema-chain metadata. The press kit notes the AI dub once, plainly. Nothing burns into the frame, nothing hampers the film, and the release sits inside Article 50 from the first screening after 2 August.

Fonti Studio renders original-cast AI dubs with the provider-side disclosure already in the file, the audio watermark and the source-file header that Article 50(2) asks of the party generating the content, plus draft consumer-disclosure language for your territory. If you have a feature heading into an EU release window and you need the dub to clear the new transparency rules cleanly, email us a brief. We will tell you what we mark and what stays yours to surface before we tell you anything about the voice.

Fonti Studio is an AI-native subtitle and dubbing service for film distributors and sales agents. €3,500 per language per dub. Flat. EUR. Free 5-minute preview before you pay.