EU AI Act Article 50 for producers: what changed on 2 August 2026

From 2 August 2026 the transparency rules in Article 50 of the EU AI Act apply, and for a film or television producer the part that matters is a single, practical duty: when an audience sees AI-generated or manipulated content that could pass for real, above all a deepfake, someone has to tell them. That obligation falls on the deployer, which in most productions is you, and it is the piece of the AI Act that touches everyday production work most directly. The Commission’s guidelines on Article 50 (opens in a new tab), published on 20 July 2026, are what turned the statute into something you can actually apply.

Provider or deployer: which one you are

The Act splits the transparency duties between two roles, and knowing which one you occupy tells you which rules are yours. A provider is the company that builds an AI system and puts it on the market; its duty is to design the system so a user knows they are dealing with a machine, and to mark the audio, image, video or text it generates in a machine-readable form. A deployer is whoever uses that system under their own authority for their own purposes, and a production company commissioning or generating content is almost always a deployer rather than a provider. The distinction decides everything that follows, so if the word is new, the short version is that you are the one using the tools, not the one that made them.

What a producer actually has to disclose

The deployer duty that reaches most productions concerns deepfakes. The Act defines one, in Article 3(60) (opens in a new tab), as AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic. If your production uses AI to create or alter footage that fits that description, you have to disclose that it is artificial, and the disclosure has to reach the viewer at the latest at the moment of first exposure, which in practice means an opening card on a video, an audible note on audio, or a visible label on an image. There is a second, narrower branch for AI-generated text published to inform the public on matters of public interest, which is more likely to bite on documentary or factual output than on drama, and it lifts where a person has taken genuine editorial responsibility for the material after a substantive review.

The carve-out that matters for creative work

The rule is not built to make you stamp a warning across an artistic choice, and the guidelines are explicit about this. Where the AI-generated or manipulated content is part of an evidently artistic, creative, satirical or fictional work, the disclosure obligation is lighter: you disclose that generated or manipulated content exists, but in a manner that does not spoil the display or enjoyment of the work, which for a feature or a scripted series usually means the credits or accompanying information rather than a label over the image. Clearly fantastical material, a dragon or a person flying unaided, sits outside the deepfake definition altogether, because nobody would take it for real. The obligation is aimed at content that trades on looking authentic, not at obvious invention.

What the Digital Omnibus did, and did not, delay

You may have heard that the AI Act was pushed back, and it is worth being precise about what moved. The Digital Omnibus, the simplification package provisionally agreed in May 2026 (opens in a new tab), postponed the deadlines for high-risk AI systems, moving standalone high-risk obligations to 2 December 2027 and those embedded in regulated products to 2 August 2028. It did not touch the Article 50 transparency duties, which took effect on 2 August 2026 as originally scheduled (opens in a new tab). The one timing concession relevant here is a transitional one: generative AI systems already on the market before 2 August 2026 have until 2 December 2026 to meet the machine-readable marking obligation, and content you published before that August date does not need retroactive labelling. So the duty to tell your audience about a deepfake is live now, whatever else was deferred.

The penalties, stated plainly

The numbers are real and it is better to know them than to guess. A breach of the Article 50 transparency obligations sits in the Act’s general penalty tier of up to €15 million or 3% of worldwide annual turnover, whichever is higher, rather than the steeper tier reserved for the outright prohibited practices. That is a ceiling set for the largest infringers, applied by national regulators with proportionality in mind, not a fixed charge that lands on any producer who gets a label wrong. It is also separate from the AI literacy duty in Article 4 (opens in a new tab), in force since February 2025, which carries no standalone fine but does expect the people using AI on your production to understand what they are working with.

The manageable starting point

Almost every part of this reduces to two questions you can answer without a lawyer in the room: which AI tools were used on the production, and where did their outputs end up in the finished work. A production that can answer those two things knows exactly which shots or tracks might need a disclosure and can decide, deliberately, how to handle them, rather than reconstructing the picture from memory months after delivery. That record is a modest thing to keep, and it is the foundation the disclosure duty rests on.

If you want the wider frame, our plain-language guide to the EU AI Act for producers sets Article 50 in context, and if you are still working out whether the deployer label applies to you, start here.