From 2 August 2026, Article 50 of the EU AI Act applies, and it places a specific obligation on anyone who publishes AI content too. A creator who posts a video made with an AI avatar or a generated voiceover becomes a deployer and shares responsibility for ensuring the viewer can recognise artificially created content.
What Article 50 requires
The regulation divides the obligation between two parties. Under Article 50(2), the provider of the system must ensure that its outputs are marked in a machine-readable way. The original text reads: “Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated.”
The mark is metadata, a watermark or a fingerprint that a detection tool can recognise even when the content is shared further.
The deployer handles the visible side. Anyone who publishes a deepfake, meaning image, audio or video content that is artificially created or manipulated and resembles a real person or event, must disclose it. The text of the regulation in paragraph 4 says: “Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake, shall disclose that the content has been artificially generated or manipulated.” The same obligation applies to artificially generated text that informs the public on matters of public interest.
A third layer concerns chatbots. A system designed to converse directly with a person must be built so that the user knows they are talking to a machine, unless this is obvious from the circumstances. For a brand that deploys an AI assistant in customer support, this is a specific requirement for the interface, not a general principle.
The provider labels, the deployer discloses
The division of responsibility changes who cannot rely on anyone else. A tool that generates a video supplies a machine-readable mark. Publication and the visible disclosure remain with the creator and the brand.
On 10 June 2026 the Commission published a voluntary Code of Practice on marking and labelling AI-generated content, which is meant to show providers and deployers a practical route to meeting their obligations.
The accompanying Commission guidelines from 20 July 2026 summarise the requirement for providers in the words “Add machine-readable marks to enable the detection of AI-generated or manipulated content.”
The rule also includes a time concession. Generative systems that were already on the market before 2 August have until 2 December 2026 to comply with machine-readable marking under Article 50(2), based on a provisional agreement on the Digital Omnibus from May 2026.
The reach does not end at the EU border. In its interpretation, the European Commission states: “Providers of AI systems established or located outside the EU are also subject to the provisions of the AI Act if the output of their AI system is used in the EU.” A US tool whose output ends up with a European audience falls under the same obligation as a European one. For a creator outside the EU who targets EU viewers, this is a reason to address provenance regardless of where they are based.
The tools creators build on
AI video platforms prepared for the deadline in advance and handle labelling through the industry standard C2PA, a signed metadata record of a file's origin. Synthesia confirmed support for the final version of the Code of Practice. Alexandru Voica, Head of Corporate Affairs and Policy, wrote in a post: “Today we're confirming Synthesia's support for the final version of the European Commission's Code of Practice on marking and labelling of AI-generated content.”
According to the same text, the videos carry “C2PA-compliant signals that identify it as AI-generated and trace the chain of custody back to its source.”
HeyGen, in its help centre, states compliance with the relevant paragraphs and puts it this way: “Complying with Chapter IV (Articles 50.2 and 50.4 of the EU AI Act), we are transparent about the AI-generated nature of content on our platform.” The company is also a member of the Content Authenticity Initiative and the C2PA coalition.
The video generator Sora builds in the same layer. In May 2024 OpenAI joined the steering committee of C2PA and committed to labelling Sora videos as AI, and its outputs carry C2PA metadata as well as a visible watermark. The provenance signal is thus moving from an optional feature to the default setting of the tools that creators actually use.
Who turns this into an advantage
The sanctions framework is high. For breaches of the transparency obligations under Article 50, the fine is up to €15 million or 3% of total worldwide annual turnover for the preceding financial year, with proportionality taken into account for small and medium-sized enterprises. How much this affects an individual creator or a smaller brand will depend on interpretation and enforcement, which are still settling.
For a creator working as a start-up, labelling AI content works as a layer of credibility. Tools such as Synthesia, HeyGen and Sora supply it as part of the price, and the audience can read it. Whoever introduces provenance and visible disclosure before a first fine forces them to do so builds a head start in a market where machine-generated content is growing quickly.
Comments