OpenAI Supports the EU Code of Practice for Trustworthy AI

⚡

Key Takeaways

1OpenAI supports the EU Code of Practice to enhance the transparency of AI-generated content, building on years of research.

2Since 2024, OpenAI has been using C2PA metadata in DALL-E 3 to improve content provenance, in collaboration with industry partners.

3The company is adopting a multi-layered approach to ensure the resilience of provenance signals, including watermarks and public verification tools.

💡Why it matters — This initiative aims to ensure responsible use of AI, which is essential for digital security and integrity in Europe.

OpenAI and the EU Code of Practice

OpenAI has recently expressed its support for the European Union's Code of Practice on Transparency of AI-Generated Content. This code, published by the European Commission, represents a crucial step in the implementation of the EU AI Act. The goal is to build a more transparent digital ecosystem where users can better understand the context of the content they encounter online. OpenAI emphasizes that this commitment is based on years of internal research and product development focused on content provenance.

History of OpenAI's Commitment

Since 2024, OpenAI has integrated C2PA metadata into its DALL-E 3 image generation tool. This initiative marks the beginning of an ongoing effort to strengthen the provenance of AI-generated content. OpenAI has developed and adopted provenance standards by adding specific metadata to its creations. This approach has been supported by active cooperation with the broader ecosystem, including hundreds of stakeholders, to ensure a trustworthy AI ecosystem.

Importance of Content Provenance

The provenance of AI-generated content is essential for users to interpret this content with confidence. Provenance signals provide information about the origin and creation of the content, helping to detect misinformation campaigns and support electoral integrity. To be effective, these methods must be developed collectively and operate interoperably across the digital ecosystem. OpenAI collaborates with coalitions like C2PA, which bring together media organizations, device manufacturers, online platforms, and AI providers to advance interoperable standards.

Technical Challenges and Solutions

C2PA metadata serves as an important foundation for provenance, allowing content to carry information about its origin, creation or modification, and the identity of those who signed this information. However, metadata is not infallible. It can be deleted, lost during downloads and uploads, or altered by transformations such as file format changes, resizing, or screenshots. To make provenance more resilient, OpenAI adopts a multi-layered approach that combines different signals, product guarantees, and collaboration across the broader ecosystem.

OpenAI's Multi-Layered Approach

OpenAI has implemented several measures to strengthen content provenance. Images generated with ChatGPT, Codex, and the OpenAI API include both C2PA metadata and SynthID watermarks. The metadata can carry richer information, while the marking can help preserve a signal across different contexts. A public verification tool is also available at openai.com/verify, allowing users to verify the provenance of images generated by OpenAI.

Contribution to Open Standards

In 2024, OpenAI joined the C2PA Steering Committee, an industry-wide effort involving software companies, camera manufacturers, media organizations, and online platforms. This participation aims to develop interoperable standards that complement policies, classifiers, reporting channels, and enforcement processes that reduce misleading uses of AI-generated content, including the misuse of likenesses and election-related misinformation.

Towards Increased Transparency

While provenance is useful, it remains an emerging field. Signals can be lost as content circulates online: metadata can be deleted, watermarks can degrade, and labels are only useful when users encounter the content. Advancing the reliability, robustness, and interoperability of provenance techniques still requires significant work and cooperation within the ecosystem. The Code of Practice sets the right direction: a high level of ambition to advance provenance techniques. In its implementation, it must remain grounded in methods that work in practice and be flexible enough to reflect the limitations of current techniques.

OpenAI will continue to enhance content transparency through its products, contribute to the development of interoperable standards, and improve verification tools. The company looks forward to collaborating with the AI Office, member states, and the broader AI community to build a safer and more transparent AI ecosystem.