The European Commission published the final Code of Practice on marking and labelling of AI-generated content. The Code is voluntary. It implements the Article 50 transparency obligations under the EU AI Act (Regulation (EU) 2024/1689), which become mandatory for providers and deployers of generative AI systems on 2 August 2026. Signatories receive streamlined compliance status across all 27 EU member states.
Article 50 of the EU AI Act imposes two obligations. Providers of AI systems that generate synthetic content must mark that content in a machine-readable format, using technical methods including watermarking where technically feasible. Deployers who use generative AI for professional purposes to produce deepfakes or AI-generated text on matters of public interest must clearly label the output for end users. The Code of Practice gives operational content to both obligations: Section 1 applies to providers; Section 2 applies to deployers. The European Commission has published a corresponding set of EU icons for use in Section 2 labelling, freely available and integral to the Code.
Generative AI system providers, including foundation model developers and application-layer companies, must integrate machine-readable content marking into their production systems before 2 August 2026. News publishers, social media platforms, advertising agencies, political communication services, and other professional deployers who use generative AI to produce content on matters of public interest face mandatory disclosure obligations at the point of publication. For signatories, the Commission intends future enforcement to focus on monitoring adherence to the Code rather than direct Article 50 proceedings, which offers material legal certainty.
Signing the Code is voluntary, but compliance with Article 50 is mandatory from 2 August 2026 regardless of whether an organisation signs. Non-signatories must demonstrate compliance through alternative means at the discretion of the relevant market surveillance authority. Technical infeasibility is an explicit qualifier for the watermarking obligation under Article 50(2), allowing providers to document infeasibility as a defence to non-compliance. The interaction of Section 2 labelling obligations with national press freedom provisions and editorial independence protections in individual member states raises questions that member state courts may need to resolve.
Licentium advises on AI Act Article 50 compliance, generative AI transparency obligations, and Code of Practice adherence strategies. We may be able to assist providers and deployers in implementing marking and labelling requirements or connect them with specialist technical and legal partners in our network. Work we undertake includes AI Act transparency compliance, generative AI system regulatory analysis, deepfake labelling policy, and multi-jurisdictional AI governance.
Source: European Commission, Code of Practice on Marking and Labelling of AI-Generated Content, 2026