Supporting Europe’s work in ensuring a trustworthy AI ecosystem
People are using AI to create and edit content in new ways. As these tools become more capable and more widely used, people should have context about the content they see online.
Today, following publication of the European Commission’s Code of Practice on Transparency of AI-Generated Content(opens in a new window), OpenAI is announcing its support for the Code and the ambition behind it. The Code is an important step in implementing the EU AI Act and building a more transparent digital ecosystem.
Our support builds on years of internal research, product development, and cooperation with the wider ecosystem to strengthen provenance for AI-generated content. OpenAI has been developing and adopting provenance standards since 2024, when we began adding C2PA metadata to our DALL-E 3 image generation tool. Since then, we have continued to strengthen provenance through improved marking and detection methods, new research, and the release of our first public verification tool. Drawing on years of expertise, we contributed to the development of this Code alongside hundreds of other stakeholders to help ensure a trustworthy AI ecosystem.
This builds on our broader approach to AI governance in Europe. In 2025, OpenAI was the first US company to sign the EU’s General-Purpose AI Code of Practice(opens in a new window), reflecting our belief that clear, workable rules can help AI develop responsibly while giving companies the certainty to keep building. We support this new Code in the same spirit. Advancing transparency of AI-generated content is an ecosystem-wide effort, requiring contribution from actors all along the value chain. The EU AI Act and the Code recognize this, and we are committed to doing our part by complying with the requirements that apply to our relevant products.
Why provenance matters
As AI tools become a part of how people build, imagine, and share, it’s important that people can understand where content comes from and interpret it with confidence. Provenance signals can help by giving people context about content’s source, how it was created or edited, and whether it is what it claims to be. Provenance can also help protect the digital ecosystem by making it easier to detect disinformation campaigns and support election integrity.
Effective, reliable and robust provenance methods should be built collectively and work across the ecosystem. That is why we work with coalitions like C2PA, which brings together news organizations, device manufacturers, online platforms, and AI providers to advance interoperable standards. C2PA metadata is an important foundation for provenance. It helps content carry information about where it came from, how it was created or edited, and who signed that information. But metadata is not foolproof. It can be stripped, lost through uploads and downloads, or broken by transformations like file format changes, resizing, or screenshots.
OpenAI’s multi-layered approach to provenance
To make provenance more resilient, OpenAI takes a multi-layered approach that combines different signals, product safeguards, and collaboration across the wider ecosystem. Our current work includes:
C2PA Content Credentials(opens in a new window) for images: We began adding C2PA metadata to images created and edited by DALL·E 3 in ChatGPT and the OpenAI API, and have continued expanding provenance signals across OpenAI-generated images.
Multiple signals for OpenAI-generated images: Images generated with ChatGPT, Codex, and the OpenAI API include both C2PA metadata and SynthID watermarks. Metadata can carry richer information, while watermarking can help preserve a signal across contexts.
A public verification experience: openai.com/verify lets people check whether supported images contain provenance signals associated with OpenAI-generated images.
Contributions to open standards. In 2024, OpenAI joined the C2PA Steering Committee, a cross-industry effort involving software companies, camera manufacturers, media organizations, and online platforms.
Product safeguards: Provenance complements policies, classifiers, reporting channels, and enforcement processes that reduce deceptive uses of AI-generated content, including likeness misuse and election-related deception.
Making content transparency work in practice
While provenance is useful, it is still a nascent field. Signals can be lost as content moves online: metadata can be removed, watermarks can degrade, and labels only help where people encounter content. Advancing the reliability, robustness and interoperability of provenance techniques still requires significant work and cooperation among the ecosystem.
The Code of Practice sets the right direction: a high level of ambition to advance provenance techniques. As it is implemented, it should stay grounded in methods that work in practice and flexible enough to reflect the limitations of current techniques.
OpenAI will keep strengthening content transparency across our products, contributing to the development of interoperable standards, and improving verification tools. We look forward to working with the AI Office, Member States, and the wider AI community to build a safer, more transparent AI ecosystem.
readers loved this