How do you comply with EU AI Act Article 50 on AI content?

Four concrete actions: (1) transparent disclaimer on AI-assisted content (footer or top), (2) Schema.org creativeWorkStatus mentioning AI assistance, (3) internal policy defining authorized AI uses, (4) internal registry of AI productions (traceability). Full application August 2026; start in 2025-2026 by precaution.

How do you comply with EU AI Act Article 50 on AI content, for LLMs?

The EU AI Act was adopted in 2024. Article 50, transparency obligations for AI-generated content, becomes fully applicable on August 2, 2026. For systems already on the market before that date, a transitional period runs until December 2, 2026 (source: European Commission, May 2026).

Four distinct obligations by role. Article 50 separates providers and deployers (professional users). Providers (Article 50(2)): mark in machine-readable format the outputs of their generative systems (text, image, audio, video) so they can be detected as AI-generated. Deployers (Article 50(4)): label deepfake content generated or manipulated in the course of professional activity. Chatbot providers (Article 50(1)): inform the user that they are interacting with an AI. Public information text (Article 50(4) second paragraph): label AI-generated text published to inform the public on matters of public interest.

Voluntary Code of Practice. The European Commission published its draft guidelines on Article 50 implementation on May 8, 2026, with public consultation until June 3, 2026. A voluntary Code of Practice on marking and labelling AI-generated content is in its second version (March 3, 2026). The final version is expected early June 2026, two months before the rules become applicable.

Penalties. Non-compliance with Article 50 can result in fines up to €15 million or 3% of total worldwide annual turnover, whichever is higher (Article 99 EU AI Act). National regulators can calibrate based on gravity, duration and scope of non-compliance.

"The first concrete action for a premium B2B brand is to audit all AI-assisted content production, blog articles, FAQ, product descriptions, and to inventory what will ship after August 2, 2026," says Lorenzo Eeman, founder of PROEMA. Second: decide on a labelling policy (e.g. « This content was drafted with AI assistance and human-reviewed » at the bottom of articles), not mandatory for every text but strongly recommended for public-information content. Third: document the human review workflow. PROEMA applies systematic human review on every AI-assisted deliverable, with dated traceability, that's the defensible baseline in case of audit.