TL;DR
This is a labeling self-check checklist for AI content creators, publishers and platforms. Under Article 50 of the EU AI Act (Regulation (EU) 2024/1689), interactive AI systems must inform users they are interacting with AI from August 2, 2026; the transition deadline for synthetic-content labelling ends December 2, 2026, by which time synthetic content systems placed on the market before August 2, 2026 must be compliant.
When to Use
Use this checklist for: teams operating chatbots and other interactive AI systems; media and creators publishing deepfakes or synthetic content; developers of generative AI apps (image, video, text); and businesses distributing AI-generated content to the EU market. Goal: meet EU AI Act transparency obligations and avoid enforcement risk after December 2, 2026.
Steps
- Determine applicability: check whether your system is an interactive AI (Article 50(1)), a deepfake (50(3)) or a synthetic content generation system (50(2)); obligations differ.
- Disclose interaction: from August 2, 2026, interactive AI systems such as chatbots must inform users at the start of interaction that they are talking to AI.
- Machine-readable marking: generative AI providers should ensure AI-generated content carries machine-readable metadata (e.g. watermarks, C2PA), making it identifiable as AI-generated.
- Visible labelling: deepfakes (image, audio, video) must be clearly and prominently labelled as AI-generated or altered; AI-generated text informing the public on matters of public interest should also be labelled.
- Cover legacy systems: for synthetic content systems placed on the market before August 2, 2026, complete Article 50(2) compliance by December 2, 2026.
- Keep records and re-check: retain version records of labelling, periodically verify that marks are not stripped or bypassed, and keep compliance evidence.
Notes
- Marks must not be easily removable: EU guidance stresses durable, traceable labelling; front-end notices alone are not full compliance.
- Three distinct duties: interaction disclosure (to users), machine-readable marking (for systems/downstream), visible labelling (for deepfakes and public-interest text); all required.
- Voluntary code helps: refer to the EU Code of Practice on Marking and Labelling of AI-generated Content and the official icon set.
- Time anchors: Article 50 core obligations apply from August 2, 2026; December 2, 2026 is the transition deadline for legacy synthetic content systems (per the official AI Act Service Desk timeline).