Transparent AI Disclosure Obligations: Who, What, When, Where, Why, How

Disclosure requirements for AI-generated content remain ambiguous: The paper identifies 149 unanswered questions about how the European AI Act's Article 52 should be implemented, revealing significant gaps between the law's intent to require transparency about AI-generated media and the practical clarity needed for compliance.

Multiple stakeholders need clearer guidance on "who" must disclose: Workshop participants highlighted confusion about which parties bear responsibility for disclosing AI use—whether it's developers, platforms distributing content, or end users—suggesting the current regulation doesn't adequately specify accountability across the AI supply chain.

"When" and "how" to disclose remain critical unsolved questions: The paper emphasizes that timing (before, during, or after content publication) and methods of disclosure (labels, watermarks, metadata) are not clearly defined, which could result in inconsistent implementation across different platforms and contexts.

Human-centered design is essential for effective transparency: The participatory approach with 16 multidisciplinary experts demonstrates that meaningful AI disclosure requires input from designers and engineers, not just legal experts, to ensure disclosures actually help users understand and evaluate AI-generated content.

Current regulations risk ineffectiveness without clearer interpretation: The paper suggests that without addressing the 149 identified questions, Article 52's transparency obligations may fail to achieve their goal of mitigating misinformation risks, as unclear requirements could lead to inadequate or inconsistent disclosure practices.