The European Commission's AI Office has opened public consultation on draft guidelines interpreting transparency obligations under the EU AI Act, marking the first formal guidance on how providers must disclose AI-generated content to users [source].

The draft guidelines address Article 50 of the AI Act, which requires providers of general-purpose AI models to ensure outputs are "marked in a machine-readable format and detectable as artificially generated or manipulated." The consultation document proposes technical standards for watermarking, metadata embedding, and user-facing disclosures across text, image, audio, and video outputs.

Key provisions under consultation include mandatory labelling thresholds, exemptions for certain use cases, and enforcement mechanisms. The guidelines propose that providers must implement "robust and reliable" technical measures within six months of the Act's application date, with specific requirements varying by modality. Text outputs would require metadata disclosure, while synthetic media would need both visible and invisible markers.

The consultation runs until 8 August 2026, inviting feedback from AI providers, civil society organisations, and member state authorities. Responses will inform the final guidelines, which the AI Office expects to publish in Q4 2026.

The transparency obligations represent one of the AI Act's most operationally significant requirements for foundation model providers. Compliance will require technical implementation across inference pipelines, potentially affecting model architectures and deployment infrastructure. The guidelines will also establish baseline expectations for how providers communicate model limitations and capabilities to end users.

The draft guidelines are available for review on the Digital Strategy portal, with submissions accepted through the online consultation form.