Written by: Haim Ravia, Dotan Hammer
The European Union has released its comprehensive Code of Practice on Transparency of AI-Generated Content, a pivotal guiding document for implementing transparency obligations under Article 50 of the EU AI Act. The Code establishes standardized rules for marking and detecting synthetic content to protect the integrity of the information ecosystem against risks of misinformation, manipulation, and fraud.
The Code divides responsibilities between AI providers and deployers. Providers of generative AI must ensure their outputs are marked in a machine-readable format and are detectable as artificially generated. Because no single technology currently meets all legal standards for effectiveness, interoperability, and robustness, the Code mandates a multi-layered approach. This typically combines digitally signed metadata with imperceptible watermarking. For free-form text longer than 200 tokens, watermarking remains a requirement, though the Code acknowledges technical reliability challenges for shorter snippets.
Furthermore, providers must make detection solutions—such as APIs or standalone software—available to the public, researchers, and authorities. These tools must be offered free of charge to regulators, fact-checkers, and civil society organizations to facilitate the identification of synthetic media. The Code also sets a deadline of February 2, 2027, for providers to implement interoperable industry standards for these detection mechanisms to ensure they work across different platforms.
For deployers of AI systems, the focus shifts to clear public disclosure. Those publishing “deepfakes” or AI-generated text intended to inform the public on matters of interest must use a new, standardized “EU icon”—a visual label featuring the acronym “AI”. This disclosure must be “clear and distinguishable” and provided at the latest at the time of first interaction or exposure. Where visual disclosure is impossible, such as in audio-only deepfakes, deployers must include a clear audible disclaimer at the start of the content.
The Code provides specific exemptions to balance transparency with other rights. Content that has undergone human review or editorial control by a media service provider is exempt from the AI-text labeling requirement. Additionally, for deepfakes used in artistic, creative, or satirical works, transparency obligations are limited to ensure the disclosure does not hamper the enjoyment of the work. Compliance will be monitored by market surveillance authorities, with simplified paths available for SMEs and startups to foster continued innovation.
Click here to read the Code of Practice on Transparency of AI-Generated Content.
