OpenAI backs EU code on AI-generated content transparency
OpenAI announced support for the European Commission's Code of Practice on Transparency of AI-Generated Content, committing to help implement standards that signal where online content originates and how it was created.
The move builds on work OpenAI began in 2024, when it started embedding C2PA metadata-a technical standard for content provenance-into images created by DALL.E 3. The company has since expanded these efforts across its image generation tools and released a public verification tool at openai.com/verify.
For product development teams, the practical challenge is clear: as AI tools generate more images and text, distinguishing authentic content from AI-created material matters for trust and platform integrity. Provenance signals help, but they're fragile. Metadata can be stripped during uploads. Watermarks degrade as images move across platforms. Screenshots break the chain entirely.
How OpenAI approaches the problem
Rather than rely on a single method, OpenAI uses multiple signals. DALL.E 3-generated images now carry both C2PA metadata and SynthID watermarks. Metadata carries richer information about creation and editing history. Watermarks persist across transformations like resizing or format changes.
The company also joined the C2PA Steering Committee in 2024, working alongside camera manufacturers, media organizations, and software platforms to build interoperable standards. This cross-industry approach matters because provenance only works if signals survive across different tools and platforms.
What product teams need to know
The EU AI Act and its Code of Practice set requirements for companies offering AI-generated content tools. Compliance means implementing detection methods, marking systems, and verification capabilities-not just adding metadata and calling it done.
For teams building with or around AI-generated images, this signals a direction: expect provenance signals to become standard practice, not optional. The standards are still evolving, and current methods have limits. But the infrastructure is moving toward making content origins traceable by default.
OpenAI said it will continue strengthening verification tools and contributing to open standards development. The company also noted that provenance works best alongside other safeguards-classifiers, reporting channels, and enforcement processes that catch deceptive uses like deepfakes or election-related disinformation.
For teams working on generative AI and LLM applications or design workflows involving AI-generated content, understanding how provenance signals work-and their current limitations-is becoming part of the job.
Your membership also unlocks:
