EU Publishes Second Draft of AI-Generated Content Labelling Code as August Deadline Looms

Europe moved closer to binding transparency rules for synthetic media on 3 May 2026, when the European Commission published the second draft of its Code of Practice on Marking and Labelling of AI-Generated Content — the voluntary soft-law instrument designed to bridge current industry practice and the hard obligations that take effect under Article 50 of the EU AI Act on 2 August 2026.

What the Draft Requires

The code carves out distinct obligations for two groups of actors. Providers — companies that develop and deploy AI systems capable of producing synthetic images, audio, or video — must embed machine-readable watermarks in their outputs. Deployers, the businesses and professionals who use those systems to create content for public distribution, bear separate disclosure duties that must be fulfilled "at the latest at the moment of first exposure."

The draft mandates a standardised visual marker — a persistent 'AI' icon — displayed continuously on deepfake video content throughout playback, not only at the beginning of a clip. This requirement addresses a documented circumvention pattern in which labels are stripped or displayed only briefly before the main content begins.

Importantly, the code applies only to lawful synthetic media. Content that violates existing law — non-consensual intimate imagery, for instance — remains subject to removal obligations under the Digital Services Act and national criminal frameworks rather than mere labelling.

Creative and satirical works receive lighter treatment. Where content is "evidently artistic, creative, satirical, or fictional," only minimal and non-intrusive disclosure is required, preserving space for parody and commentary while closing the loophole that would allow misleading but stylistically ambiguous content to avoid labelling altogether.

Private individuals creating AI content for personal use are exempt from the code's obligations entirely.

The Regulatory Timeline

The first draft of the code appeared on 17 December 2025. The Commission intends to finalise the document in May or June 2026, giving industry several weeks to align internal processes before Article 50 becomes enforceable. Companies that participate in drafting the code and subsequently fail to comply once it is finalised face reputational and, eventually, legal risk — the AI Act empowers national market surveillance authorities to investigate and penalise non-compliance from August onwards.

The code operates at the intersection of three separate EU legal instruments: the AI Act itself sets the transparency baseline for generative AI providers; the Digital Services Act imposes due-diligence obligations on very large online platforms that host AI-generated content; and GDPR's data-protection principles constrain how biometric data processed during deepfake detection and watermark verification can be stored and shared. Compliance teams at large platforms must coordinate across all three regimes simultaneously.

Why This Matters Now

Deepfake video has proliferated rapidly since 2024, with documented misuse in political advertising, financial fraud, and non-consensual pornography. The Commission's push for a finalised code before August reflects awareness that statutory text alone is insufficient: Article 50 establishes the obligation but does not specify technical standards. The code fills that gap by anchoring the legal duty to concrete, auditable practices such as C2PA-compatible watermarking and the persistent icon requirement.

Several large platforms — including video-sharing services and social networks — have already embedded preliminary watermarking capabilities in response to earlier drafts. Finalisation of the second draft is expected to trigger a wave of conformance testing and, in some cases, product changes to satisfy the continuous-display requirement for the AI icon.

The code also signals how the Commission plans to govern other categories of AI-generated content in the future. A modular structure in the draft explicitly envisions annexes covering audio deepfakes and text-only synthetic content, suggesting that the August 2026 visual-media regime is the first layer of a broader transparency architecture rather than a standalone measure.

What to Watch

The key variable is whether the May 13 Digital Omnibus trilogue — which covers a separate set of AI Act amendments — produces a deal that could alter Article 50's scope or timeline. If the Omnibus fails, Article 50 stands as currently drafted and the code's provisions become the operative compliance standard for every organisation producing or hosting AI video content in the EU.