European Commission Drafts AI Transparency Code

The European Commission has published the first draft of the AI Code of Practice on transparency, intended to support compliance with the EU Artificial Intelligence Act. The Code is voluntary and functions as non-binding guidance, allowing companies to opt in while demonstrating good‑faith alignment with the Act’s transparency obligations.

A core focus of the draft Code is the disclosure of photorealistic AI‑generated content, including images and videos that appear to depict real persons or events. For such deepfakes, the drafters propose the creation of a common EU icon that would signal at a glance that content has been generated or modified using AI and provide access to additional information.

The icon is expected to include a two‑letter acronym referring to artificial intelligence, most likely “AI,” with adaptations possible for different EU languages. Signatories would commit to supporting the development of this common symbol and, in the interim, to using a simplified version consisting solely of the acronym until a final design is adopted.

Beyond deepfakes, the draft Code requires companies to provide provenance information for other AI‑generated content, preferably through metadata combined with an imperceptible watermark embedded directly in the content. Companies are expected to use technically effective and economically viable watermarking methods, and model developers would commit to training AI systems in ways that enable downstream users to clearly identify synthetic outputs.