Marking Obligation Grace Periods Expires on December 2, 2026

Article 50(2) requires providers of AI systems that generate synthetic audio, image, video, or text content to ensure their outputs are marked in a machine-readable format, detectable as artificially generated. This obligation, along with the rest of Article 50, became applicable on August 2, 2026.

A narrower transitional allowance was granted specifically for this marking requirement as part of the Digital Omnibus package: providers of AI systems already placed on the market before August 2, 2026 were given until December 2, 2026 to bring existing systems into compliance. That grace period expires on this date.

This is worth distinguishing clearly from a separate, much larger deferral in the same Digital Omnibus package: the applicability of new high-risk AI system obligations under Annex III was pushed back to December 2027. That deferral does not apply to Article 50, yet some public commentary has conflated the two, creating a mistaken impression that Article 50 obligations broadly were delayed. They were not, only this specific transitional window for pre-existing marking systems was extended, and only until this date.

It’s also worth being precise about what this deadline does and doesn’t cover. Article 50(2) is a provider obligation, it requires the company building the AI system to mark its output. It says nothing about a deployer’s separate duty under Article 50(4) to disclose AI-generated text published on matters of public interest, unless that content underwent genuine human review with editorial responsibility. A provider’s system being properly marked after this date does not, on its own, satisfy a publisher’s own disclosure obligations under the separate provision.

We’ll update this entry if the Commission or any national market surveillance authority issues further guidance or takes enforcement action tied to this deadline.