EU AI Act Transparency Obligations: Code of Practice and Draft Guidelines

The European Union is taking significant strides in regulating artificial intelligence with the recent publication of its Code of Practice on Transparency of AI-Generated Content. This comprehensive framework, finalized on June 10, 2026, serves as a crucial guide for organizations developing and deploying AI systems, particularly in light of the transparency mandates within Article 50 of the EU AI Act. A detailed analysis from Mishcon de Reya, published on June 25, 2026, sheds light on the practical implications of these new guidelines for the industry. The primary objective of the Code is to ensure that AI-generated or manipulated content is clearly identifiable. It advocates for a multi-layered marking approach, especially in scenarios where a single detection method might not suffice to guarantee effectiveness, interoperability, robustness, and reliability. This often necessitates the implementation of at least two distinct machine-readable marking layers. For instance, the Code specifies that free-form text exceeding 200 tokens must be watermarked, though exceptions are made for shorter text segments and AI systems integrated into controlled physical products. While the Code of Practice itself is voluntary, the core transparency obligations embedded in Article 50 of the EU AI Act are legally binding and are slated to become enforceable on August 2, 2026. However, a transitional period has been introduced for certain provider obligations concerning the marking and detection of AI systems that were already on the market before this date, pushing their compliance deadline to December 2, 2026. In contrast, deployer obligations are not afforded this grace period and will commence as scheduled on August 2, 2026. The Code outlines several key commitments for providers, including the integration of content marking "by design" into their models, the implementation of measures to preserve these markings against tampering, and the provision of accessible detection solutions, ideally free of charge and hosted within the EU. For deployers, the emphasis is on transparent public disclosure. This means that "deepfakes" and any AI-generated text intended for public consumption must be clearly labeled with a standardized "EU icon". This regulatory push is designed to build public trust and counter the spread of misinformation by enabling users to easily differentiate between human-created and AI-generated content in the evolving digital landscape.