Advancing Transparency of AI-Generated Media in the EU Code of Practice

2026-05-06 · Source: Partnership on AI · Field: Government & Public Sector — Public Policy & Governance, Artificial Intelligence & Machine Learning · Depth: Intermediate, medium

Summary

The Partnership on AI (PAI) has submitted recommendations to the EU Code of Practice on AI-Generated Content, aiming to enhance transparency in synthetic media. This initiative addresses the growing challenge of distinguishing authentic from AI-generated content, which erodes trust in information and has been used for election manipulation, fake profiles, and misrepresentation in conflicts. PAI advocates for a multi-layered transparency ecosystem that combines robust technical markings, such as watermarking, fingerprinting, and cryptographic metadata, with clear user-facing disclosures. The recommendations also emphasize balancing transparency with security for detection technologies, standardizing direct disclosure icons, creating a repository for creative content disclosure case studies, and investing in user research and education to ensure transparency signals are effective and understood across diverse demographic groups. PAI also expressed concern over the removal of model-level transparency from the draft Code.

Key takeaway

For CTOs and VPs of Engineering developing or deploying generative AI systems, prioritizing a multi-layered transparency framework is crucial. Your teams should integrate robust technical markings like watermarking and cryptographic metadata, alongside standardized, user-friendly direct disclosure icons. Invest in user research to validate the effectiveness of these signals across diverse demographics, ensuring your transparency efforts genuinely build trust and do not create a false sense of security for consumers.

Key insights

Effective synthetic media transparency requires a multi-layered approach combining technical markings with clear user-facing disclosures.

Principles

• Transparency must preserve privacy and support innovation.
• Detection accuracy is never 100% and can become outdated quickly.
• Inconsistent terminology limits transparency efficacy.

Method

Implement three layers of technical marking (watermarking, fingerprinting, cryptographic metadata) and standardize direct disclosure icons, supported by user research and education.

In practice

• Use watermarking, fingerprinting, and cryptographic metadata.
• Standardize direct disclosure icons for user clarity.
• Conduct user research on transparency signal impact.

Topics

• AI-Generated Media Transparency
• EU Code of Practice
• Synthetic Media Framework
• Technical Marking
• Detection Technologies

Best for: CTO, VP of Engineering/Data, Director of AI/ML, Policy Maker, AI Ethicist, Consultant

Related on AIssential

• Supporting Europe’s work in ensuring a trustworthy AI ecosystem — Effective AI content provenance requires multi-layered technical approaches and broad ecosystem collaboration to ensure transparency.
• AI Disclosure Labels Risk Becoming Digital Background Noise — Over-labeling AI content risks user habituation, making disclosures ineffective during critical events.
• Shaping AI Transparency Processes with NIST — Effective AI governance requires balanced, accessible, and iteratively developed documentation standards.
• Thoughts on Watermarking AI-Generated Content — Watermarking AI-generated content is technically feasible and robust enough for practical use, but requires broad adoption and regulatory coordination to address societal problems.
• Authenticity Debt and the Synthetic Content Threat Landscape: A Layered Framework for Trust, Provenance, and IP Governance in the Generative AI Era — Authenticity debt accumulates when AI-generated content lacks verifiable origin, necessitating a layered trust framework.
• The Download: Microsoft’s online reality check, and the worrying rise in measles cases — Microsoft proposes technical standards to verify online content authenticity against advanced AI deception.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Partnership on AI.