What happened

The transparency obligations in Article 50 of the EU Artificial Intelligence Act (Regulation (EU) 2024/1689) became applicable on 2 August 2026. According to the European Commission, providers and deployers of AI systems within the scope of Article 50(2) and (4) must comply with the relevant transparency obligations for AI-generated or manipulated content from that date.

To help organisations comply, the Commission published a voluntary Code of Practice on Transparency of AI-Generated Content in June 2026, drawn up by independent experts, and guidelines explaining the scope of the Article 50 obligations. Adhering to the code is voluntary, but the Commission stresses that the transparency requirements themselves are legal obligations.

The date that matters next is 2 December 2026. Legal analyses of the EU's Digital Omnibus changes, including those by Paul, Weiss and Bristows, describe a grace period until that date for the machine-readable marking and detection obligations that apply to generative AI systems already on the market before 2 August 2026. Systems placed on the market on or after 2 August 2026 had to comply immediately.

What the rules require

Article 50 covers several distinct situations rather than one universal "AI" badge:

  • Talking to an AI. People should be told when they are interacting with an AI system, unless it is obvious.
  • Marking synthetic content. Providers of generative AI systems must ensure that audio, images, video and text generated by their systems are marked in a machine-readable way and are detectable as AI-generated or manipulated. This is the obligation with the December grace period for older systems.
  • Deepfakes and public-interest text. Deployers who publish deepfake images, audio or video, or AI-generated text intended to inform the public on matters of public interest, must disclose that it was artificially generated or manipulated, subject to exceptions set out in the Act.
  • Emotion recognition and biometric categorisation. People exposed to these systems must be informed.

The Code of Practice gives signatories a recognised, practical route to demonstrate compliance with the marking and labelling parts. Techniques discussed in this area include metadata, watermarks and visible labels; the code also encourages optional measures such as provenance information in metadata.

What it means for everyday users

For most people, the change will appear gradually as more visible labels and "made with AI" notices on platforms, and as more AI-generated files carrying hidden provenance information such as Content Credentials or watermarks. You are not required to label an AI picture you make for a private chat, but organisations publishing deepfakes or public-interest AI text in the EU now have disclosure duties.

There is a practical side effect worth knowing. Machine-readable marks stored in metadata can be lost when a file is re-encoded, screenshotted or uploaded to a service that strips metadata. Convertios tools that re-encode images, such as the image compressor and the Image Converter, create new files and do not carry over the original metadata. That is helpful for removing location data, but it also removes provenance metadata. If you publish AI-generated material where labelling matters, keep the original file and add any required visible disclosure in the place where you publish. Convertios does not add, read or verify AI-content labels.

What to check

  • Publishing in the EU? If you deploy deepfakes or AI-generated text on matters of public interest, review the Commission's guidelines and your disclosure wording.
  • Building or reselling a generative AI tool? Check whether the marking and detection obligations apply to you and whether the December transition applies to your system.
  • Editing AI images? Use tools that preserve provenance data when it matters, and keep originals.
  • Reading AI labels? Treat a label as useful information about origin, not proof that content is true or false.

Limits and status

This article is general information, not legal advice. The Commission's guidance and the code explain how the obligations can be met, but how they apply to a specific organisation depends on its role (provider or deployer), its systems and the exceptions in the Act. Dates for specific obligations were adjusted by later EU legislation; rely on the Commission's current pages and professional advice for compliance decisions. Machine-readable marking is also not a complete defence against misuse: metadata can be removed and watermarks can be weakened, which is why the rules combine technical marking with human-readable disclosure.