OpenAI Deploys C2PA and SynthID to Authenticate AI Images
OpenAI has integrated C2PA metadata and SynthID watermarking into images generated with ChatGPT, Codex, and its API, marking a major step in AI image provenance. C2PA, developed by the Coalition for Content Provenance and Authenticity, attaches cryptographic metadata to files detailing creation and edits. SynthID, created by Google, embeds an invisible watermark directly into the image, surviving screenshots and manipulation. Both systems have limitations: C2PA metadata can be removed or tampered with, while SynthID is more resilient but less informative. A recent photo competition winner was disqualified after a SynthID watermark was detected. OpenAI also launched a public 'Verify' tool allowing anyone to upload an image and check for OpenAI-originated watermarks or C2PA manifests. OpenAI states errors are rare. The deployment by a leading AI firm is seen as a necessary response to the proliferation of AI-generated images and videos online.
Key facts
- OpenAI adds C2PA metadata and SynthID watermark to images from ChatGPT, Codex, and API.
- C2PA is a metadata standard by the Coalition for Content Provenance and Authenticity with cryptographic verification.
- SynthID is an invisible watermarking technology by Google embedded in the image itself.
- C2PA metadata can be tampered with or removed; SynthID is more resilient to screenshots and manipulation.
- A photo competition winner was disqualified due to a SynthID watermark on their image.
- OpenAI launched a public 'Verify' tool to check if an image originated from OpenAI tools.
- The Verify tool looks for SynthID watermarks or trusted C2PA manifests from OpenAI.
- OpenAI notes that errors in verification are rare.
Entities
Institutions
- OpenAI
- Coalition for Content Provenance and Authenticity
- Coalition for Content Provenance and Authenticity (C2PA)