Machine-readable marking of generated content is now enforceable
Under Article 50, providers of systems generating synthetic audio, image, video or text must ensure outputs are marked in a machine-readable format and detectable as AI-generated. Systems already on the market have until 2 December to comply with the marking requirement specifically.
The staggered deadline is the regulator conceding that this is an engineering problem rather than a policy one. Telling a user they are speaking to a chatbot is a disclosure string. Making every generated output detectable in a format that survives re-encoding, cropping and platform processing is a provenance pipeline.
Audio is the least developed case and it is now in scope alongside images. Resampling, compression and re-recording degrade a signal differently from how cropping degrades pixels, and the watermarking literature is far thinner there.
Models that generate synchronised audio and video jointly raise a question with no settled answer: one mark or two. Mark separately and an attacker strips one channel; mark jointly and the mark must survive two processing pipelines that are rarely applied together.
EU Artificial Intelligence Act — The EU AI Act's transparency rules: a practical guide to Article 50 → · Pearl Cohen — New guidance under the EU AI Act ahead of its next enforcement date → · Cooley — EU AI Act: transparency obligations take effect 2 August 2026 →