The EU AI Act for film and TV producers: a plain-language guide

The EU AI Act (opens in a new tab) is the first broad law to regulate artificial intelligence, and for a film or television producer the honest summary is that most of it is someone else’s problem, with two exceptions that apply to you now: telling your audience when they are seeing AI-generated content, and making sure the people using AI on your production understand what they are working with. The Act is built around risk, it treats a medical-triage algorithm very differently from a tool that upscales a shot, and the kind of AI a production actually uses sits at the lighter end of that scale. This guide is the plain-language version of where you sit, what applies when, and what you can safely set aside.

How the Act sorts AI, and where your work lands

The Act splits AI into tiers by the risk it poses, and the tier decides the obligations. A small set of uses is prohibited outright, things like social scoring or manipulative systems that exploit vulnerability, and these have been banned since February 2025. A larger category is high-risk, covering AI that makes or heavily informs consequential decisions about people, such as in recruitment, credit, education or biometric identification, and this is the tier that carries the demanding obligations most coverage of the Act describes. Below that sits a category the Act handles through transparency rather than heavy control, where the concern is not that the system decides your fate but that you might not realise you are dealing with AI at all, and this is where the tools a production uses to generate images, audio, video and text almost always fall. There is also a minimal-risk remainder that the Act largely leaves alone.

The practical consequence is worth stating plainly, because it is the single most useful thing a producer can take from the Act: your generative tools are, in the ordinary case, limited-risk systems governed by the transparency rules, not high-risk systems governed by the full compliance regime. The heavy obligations you may have read about, the conformity assessments, the risk-management systems, the registration in an EU database, attach to the high-risk tier, and a production that is using AI to draft a logline, clean up a plate or dub an episode is not operating in it.

The dates that matter to a producer

The Act came into force in 2024 and switched on in stages. The prohibitions and a duty to build AI literacy took effect on 2 February 2025. Rules for general-purpose AI models, the large foundation systems, followed on 2 August 2025 and bind the companies that make those models rather than you. The date that matters most to a production is 2 August 2026, when the Article 50 transparency obligations began to apply, and a short transitional window runs to 2 December 2026 for the machine-readable marking of generative systems that were already on the market before that August date.

The one recent change worth understanding is what the Digital Omnibus, a simplification package that entered into force on 27 July 2026 (opens in a new tab), actually did. It pushed the high-risk deadlines back, moving standalone high-risk obligations to 2 December 2027 and those embedded in regulated products to 2 August 2028, and it left the Article 50 transparency duties exactly where they were. So if you took away one impression from the reporting that the Act had been delayed, the part that reaches you was not the part that moved.

The two duties that are live for you now

The first is disclosure. Under Article 50 your production has to tell audiences when they are seeing AI-generated or manipulated content that could pass for real, above all a deepfake, and the disclosure has to reach the viewer at first exposure. There is a lighter version of this obligation for evidently artistic or fictional work, and the what-changed-on-2-August guide sets out how it works in practice and where the deepfake definition begins and ends. This is the duty that touches everyday production most directly.

The second is quieter and older, and it is easy to miss because it carries no headline fine. Since February 2025 the Act has expected providers and deployers to ensure a level of AI literacy among the people who use these systems on their behalf, which for a production means the crew and contractors using AI tools should understand, at a working level, what those tools do and where their limits are. The Digital Omnibus simplified how this duty is framed rather than removing it. It is not a certification exercise, and treating it as basic professional competence rather than a compliance project is closer to the spirit of the rule.

When a production could be more than a limited-risk deployer

The limited-risk framing holds for the AI most productions touch, but it is worth knowing the edge cases so you can recognise one if it appears. If your company uses an AI system to screen job applicants, to score or rank the people you hire, or to run biometric identification, you may have stepped into the high-risk tier, and the fuller set of deployer obligations would then apply to that specific use. That is a decision about people rather than a step in making the picture, and it is uncommon on a production, but a growing production company with its own hiring tools is exactly the kind of organisation that could acquire a high-risk deployment without noticing. The point is not to worry about it in the abstract; it is to know which of your uses might cross the line, so you can look at that one deliberately.

What the Act is not

It is not a ban on using AI in production, and nothing in it tells you that you cannot generate, edit or localise with these tools. It does not ask you to prove how much time or money or carbon AI saved you, because it is not a measurement regime; it is a transparency and safety regime. And it is not retroactive in the way that would force you to relabel a back catalogue, since content published before the transparency rules applied does not need to be marked after the fact. Reading the Act as a set of narrow, mostly manageable duties rather than an existential threat is not optimism, it is an accurate reading of how the tiers are drawn.

The spine that makes all of it manageable

Almost everything the Act asks of a production rests on one modest habit: knowing which AI tools were used on a production and where their outputs ended up in the finished work. A production that can answer those two questions can tell which content might need a disclosure, can brief its people on the tools they are actually using, and can hand a broadcaster or an insurer a straight account instead of a shrug. That record is the foundation the rest of this cluster builds on, and it is a great deal less work to keep as you go than to reconstruct after delivery.

From here, the natural next steps are the posts this guide links down to: what changed on 2 August 2026 if you want the disclosure duty in detail, are you a deployer if the roles are still unclear, and what non-compliance actually costs if you want the penalties stated plainly rather than brandished.