EU AI Act Article 50: what labelling obligations before 2 August 2026?

Article 50 of Regulation (EU) 2024/1689 (EU AI Act) imposes from 2 August 2026 several transparency duties: label AI-generated content, flag interactions with AI chatbots, and identify deepfakes. Theoretical penalty: €15M or 3% of global turnover. Take this to a lawyer, at PROEMA we consult Sabrina before any implementation.

EU AI Act Article 50: what labelling obligations before 2 August 2026, for a brand?

The EU AI Act timeline is not a suggestion: six dates structure a countdown whose final stage, August 2, 2026, triggers effective enforcement of Article 50 transparency obligations.

August 2, 2024, Official publication of the EU AI Act in the Official Journal. Entry into force twenty days later but with phased application. March 3, 2026, European Commission publishes the second version of the voluntary Code of Practice on marking and labelling AI-generated content. May 8, 2026, Draft guidelines for Article 50 implementation released, with open public consultation. June 3, 2026, End of public consultation on the guidelines. Early June 2026, Expected finalisation of the Code of Practice (confirmed by the Commission). August 2, 2026, Effective enforcement of Article 50 for new systems placed on the market. December 2, 2026, End of the transitional period for systems already on the market before August 2, 2026.

Four types of labelling obligations. Type 1 (Article 50(1), providers): inform the user of a chatbot, conversational agent or AI assistant that they are interacting with a machine. Type 2 (Article 50(2), providers): mark in machine-readable format AI-generated outputs (text, image, audio, video) to enable automated detection. Primarily aimed at provenance and authentication. Type 3 (Article 50(4), deployers): label deepfakes (artificially generated or manipulated visual/audio content resembling real people or events) distributed professionally. Type 4 (Article 50(4) second paragraph, deployers): label AI-generated text published to inform the public on matters of public interest.

Penalty threshold. Article 99 sets fines up to €15M or 3% of total worldwide annual turnover, whichever is higher, for Article 50 breaches. For an SMB, that's potentially existential; for a group, a provisioning line.

What to do before August 2026? Three immediate actions: audit AI-assisted content shipped over the last 12 months, choose a labelling policy (formal or contextual), document the human review workflow. "Systematic human review remains the strongest defence if you are audited, and it is the standard we apply to every AI-assisted editorial production," says Lorenzo Eeman, founder of PROEMA.