AI watermarking and Article 50 of the EU AI Act

Article 50 of the EU AI Act requires providers of generative AI systems to make certain synthetic or manipulated outputs detectable through machine-readable marking. It came into force on 2 August 2026.

The requirement does not apply where an AI system performs an assistive function for standard editing or does not substantially alter the user’s input or its meaning. However, some providers’ implementations may mark content that has been proofread, translated or otherwise processed by an AI system.

This creates practical questions for organisations and individuals using AI to write, edit and review content. A mark may show that an AI system processed material, but it does not necessarily establish who wrote it, who owns it or how substantially AI contributed to it.

In this webinar, we will cover:

• What Article 50 requires and the circumstances in which its marking requirement does not apply
• The distinction between AI-generated and AI-assisted content
• How providers are implementing machine-readable marking
• What a watermark can—and cannot—prove
• Whether marks survive editing, copying, translation, middleware and deliberate attempts to remove them
• The reliability of detection, including false positives, false assumptions and missing marks Reputational risks for creators, professionals and businesses
• What provenance metadata may reveal about the system used, processing date or extent of AI involvement
• The associated privacy, confidentiality and data-protection risks
• Whether watermarking will change working practices or shift users towards providers that take a different approach
• How organisations can adopt proportionate policies, comply with Article 50 and protect trust in human work