OpenAI backs EU AI content transparency code with C2PA and SynthID

AI content labeling is moving from product polish into regulatory plumbing. On June 11, 2026, OpenAI said it supports the EU Code of Practice on Transparency of AI-Generated Content, framing provenance as part of how the EU AI Act becomes workable in real products.

The concrete change is not a new model. It is the stack around generated media. In its official post, OpenAI says it began adding C2PA metadata to DALL-E 3 images in 2024 and has since expanded provenance signals across images created or edited through ChatGPT, Codex, and the OpenAI API. Those images can carry both C2PA Content Credentials and SynthID watermarks.

That layered approach matters because metadata alone is fragile. OpenAI notes that provenance data can be stripped or broken by ordinary online handling such as uploads, downloads, format changes, resizing, or screenshots. C2PA can carry richer signed information about origin and editing history, while watermarking can preserve a weaker but more resilient signal when metadata disappears.

The company also points to a public verification experience at openai.com/verify for supported images. That is a small detail with a larger implication: provenance only helps if downstream users, platforms, journalists, and investigators have a way to check the signal, not merely trust that a label was attached somewhere upstream.

For builders, the EU code is a reminder that AI-generated media is becoming an ecosystem problem. Providers need metadata, watermarks, policy enforcement, reporting channels, and verification tools to work together. Platforms need to preserve and display those signals. Regulators need methods that survive the messy reality of screenshots and reposts. The next fight is less about whether AI content should be labeled, and more about which signals still mean anything after the internet has handled the file.