The transparency guidance details disclosure standards, identification of AI works, member-state enforcement rights, and a list of exemptions.
Just ahead of the 2 August 2026 compliance deadline, the European Commission has issued its final guidance on transparency requirements under the EU AI Act, setting the stage for enforcement.
The guidance focuses particularly on obligations tied to AI-generated content, including deepfakes, and clarifies how both developers and deployers of AI systems are expected to meet the regulation’s disclosure standards.
Central to the rules is Article 50, which outlines four distinct categories of transparency obligations:
- Providers of AI systems must ensure that users are clearly informed when they are interacting with artificial intelligence, such as in the case of chatbots or virtual assistants.
- Developers of generative AI tools are required to incorporate machine-readable identifiers into synthetic outputs—spanning text, audio, images, and video — so that such content can be reliably detected as AI-generated.
- The regulation also places responsibilities on those deploying AI systems. Organizations must explicitly disclose when content qualifies as a deepfake, defined broadly as synthetic or manipulated media that could be mistaken for authentic material.
- Finally, any AI-generated text disseminated to inform the public on issues of public interest must be clearly labeled. Transparency obligations extend further to systems involving emotion recognition or biometric categorization, where individuals must be informed when they are subject to such processing
At the same time, the European Commission’s guidance draws boundaries around what constitutes a deepfake. Minor edits to existing content — such as adjustments to lighting, color balance, or background elements — may fall outside the scope of labeling requirements.
Similarly, content that is clearly artistic, fictional, or satirical is subject to lighter disclosure obligations, provided that transparency measures do not undermine the intended user experience.
Heavy penalties for non-compliance
While the guidelines are not meant to create new legal obligations on their own, they are expected to play a significant role in how national regulators and courts interpret compliance under the AI Act. Enforcement authority will rest with market surveillance bodies in each EU member state, which will gain the legal powers starting 2 August 2026. Non-compliance could result in penalties of up to €15m or 3% of a firm’s global annual revenue.
A limited transition period has been granted for certain technical requirements. AI systems already available before the August deadline will have until 2 December 2026, to implement watermarking and detection features, following a political agreement reached earlier this year.
To complement the regulatory framework, a voluntary Code of Practice on AI-generated content transparency was finalized in June 2026 by an independent expert group. The code proposes practical measures such as standardized labeling icons and a layered approach that combines visible disclosures with embedded metadata and watermarking techniques. Firms that adopt these practices may use adherence to the code as evidence of compliance with the AI Act.
