OpenAI joins C2PA and adds Google SynthID watermarks to provenance stack

TL;DR:

  • OpenAI has become a C2PA Conforming Generator Product, allowing other platforms to read, preserve and pass along the cryptographic provenance metadata OpenAI attaches to generated content.
  • The company has integrated Google DeepMind’s SynthID invisible watermarking into images generated by ChatGPT, Codex and the OpenAI API, adding a second durable layer of provenance signal.
  • A public verification tool is now in preview, letting anyone upload an image and check whether it was generated using OpenAI tools — initially limited to OpenAI content, with plans to expand cross-industry.

The combination — open-standard metadata, durable watermarking and consumer-facing verification — is the most complete provenance stack any major AI lab has shipped publicly. OpenAI has supported C2PA since 2024 when it began adding Content Credentials to DALL·E 3 images, and joined the C2PA steering committee that year; the new step is formal Conforming Generator certification.

A multi-layered approach

OpenAI’s framing is that no single technique is enough. C2PA metadata carries detailed signed information about where content came from, who created it and how it was edited — but metadata can be stripped, lost in re-uploads, or broken by format conversion. SynthID embeds an invisible pixel-level signal that survives transformations like screenshots and format changes that metadata typically does not. The two systems reinforce each other: C2PA carries detailed context where it survives; SynthID provides a more durable but less detailed fallback signal.

The public verification tool — at openai.com/verify — handles cases where either signal is detected, including cases where C2PA metadata has been stripped but a SynthID watermark remains. OpenAI says no detection method is foolproof, and the tool will not make definitive claims when both signals are absent. At launch, verification is limited to content OpenAI generated; cross-industry verification (where the tool could verify Anthropic, Google or other lab-generated content) is on the roadmap, dependent on those labs adopting compatible standards.

Looking forward

For UK readers, this development sits squarely inside an active policy debate. Ofcom’s Online Safety Act enforcement programme has been weighing how to handle AI-generated content on regulated platforms, particularly synthetic explicit imagery and deepfake political content. The Information Commissioner’s Office has flagged AI-content provenance as an emerging area for guidance. The C2PA standard has political traction at the EU level via the AI Act, and the UK has informally aligned via the AI Safety Institute and DSIT. The harder question is whether OpenAI’s verification tool — currently a free public preview limited to OpenAI content — becomes the de facto consumer-facing UI for AI-content authenticity, or whether a platform-neutral verifier emerges from industry collaboration. The integration with Google DeepMind’s SynthID is a useful signal that the bigger labs see provenance as collaborative infrastructure rather than competitive differentiation.

Share this article