When does the EU AI Act labelling rule start?

The Article 50 transparency obligations apply from 2 August 2026. Tool makers whose generative systems were on the market before then have until 2 December 2026 for machine-readable marking.

Do I have to label every AI video in the EU?

The legal duty under Article 50(4) is for deepfakes made in a professional capacity. Platforms still require labels on realistic AI content regardless.

Does a watermark from the tool count as my disclosure?

The provider's machine-readable marking is a separate duty from your disclosure. For deepfakes, add a disclosure viewers can see or hear.

Do I need to relabel videos I posted before August 2026?

No, the Commission says earlier content need not be labelled retroactively, though it encourages it.

Do AI-generated videos need a label under the EU AI Act?

Yes, if the video is a deepfake made in a professional context: from 2 August 2026, Article 50(4) of the EU AI Act requires you to disclose that AI video resembling existing people, places or events, which would falsely appear authentic, was artificially generated^[1]^[5]. Separately, the companies that make AI video tools must mark every output in a machine-readable way^[2].

This page is general information, not legal advice. It summarises the regulation text and the European Commission's own FAQ; how a national regulator applies them to your case may differ.

Two different duties in the transparency rules

Who must do what under the EU AI Act's transparency rules.
Duty	Who it applies to	What it requires	From
Machine-readable marking, Article 50(2)^[2]	Providers of generative AI systems (the tool makers)	Mark outputs in a machine-readable format detectable as AI-generated	2 August 2026; 2 December 2026 for systems already on the market before 2 August 2026^[7]
Deepfake disclosure, Article 50(4)^[1]	Deployers: people or businesses using the AI system in a professional capacity	Disclose that deepfake image, audio or video content was artificially generated or manipulated	2 August 2026^[5]

The extension to 2 December 2026 comes from the AI Omnibus, Regulation (EU) 2026/1744, and covers only the marking duty for generative systems that were already on the market^[7]. It does not move the deepfake disclosure date.

Is your video a deepfake?

The Act defines a deep fake as AI-generated or manipulated image, audio or video that resembles existing persons, objects, places, entities or events and would falsely appear authentic to a person^[3]. Read against that definition:

A photoreal clip of a real politician saying something they never said is a deepfake.
A realistic street scene of a real city during an event that never happened can be one.
A clearly stylised cartoon character in an invented world is much less likely to be one, because it would not falsely appear authentic.
A photoreal invented person sits in between: the definition talks about resembling existing persons, so how realistic and how believable the scene is will matter.

For evidently artistic, creative, satirical or fictional work, the duty is limited to disclosing the AI content in a way that does not hamper enjoyment of the work^[4]. Our reading: that points to a disclosure that sits alongside the work, such as a caption or a credit, rather than one that interrupts it.

Does it apply to creators?

The Commission's FAQ says purely personal, non-professional use, such as making deepfakes for your own social media, is outside the AI Act, but regular economic activity makes you a deployer^[9]. If you run a monetised channel, make videos for clients or post as a business, assume the disclosure duty applies to you.

Content generated before 2 August 2026 does not have to be labelled retroactively, though the Commission encourages it^[8].
The Commission's Code of Practice on transparency of AI-generated content is voluntary, but the Article 50 requirements themselves are legal obligations^[6].

Platform labels are a separate rule

Platforms ask for labels whatever the law says. TikTok requires a label on AI content with realistic images, audio and video^[10], YouTube requires disclosure of realistic altered or synthetic content^[12], and Meta requires its disclosure tool for photorealistic AI video^[13]. Platforms can also read machine-readable markers such as C2PA and label your upload automatically^[11]. Step-by-step: how to label AI videos on TikTok and disclosing AI videos on YouTube.

Where Cast fits

Stylised or clearly fictional characters, where a light caption-level disclosure is the most the artistic exception asks for.
Creators who want to label honestly: Cast adds no visible watermark of its own, so you add the label the platform asks for.

Where it does not:

Anything built on a real person's likeness: Cast cannot make it, and it is the case the deepfake rule targets.
Teams that need provider-side compliance documentation for the AI Act marking duty: Cast has not published any.

Make a video · first video free

Sources

Claims about other companies come from their own pages, read on the date shown. Products and prices change; check the link for the current version.

EU AI Act (Regulation (EU) 2024/1689), Article 50(4) (read 2026-10-18)
EU AI Act, Article 50(2) (read 2026-10-18)
EU AI Act, definition of deep fake (read 2026-10-18)
EU AI Act, Article 50(4), artistic and fictional works (read 2026-10-18)
European Commission: Code of Practice on AI-generated content (read 2026-10-18)
European Commission: Code of Practice, legal status (read 2026-10-18)
Regulation (EU) 2026/1744 (AI Omnibus) (read 2026-10-18)
European Commission: FAQ on Article 50 transparency obligations (read 2026-10-18)
European Commission: FAQ on Article 50, personal use (read 2026-10-18)
TikTok: AI-generated content labels (read 2026-10-18)
TikTok: automatic AI labels and C2PA (read 2026-10-18)
YouTube Help: disclosing altered or synthetic content (read 2026-10-18)
Meta: labeling AI-generated images on Facebook, Instagram and Threads (read 2026-10-18)