OpenAI Adopts Google SynthID Watermarks and Verification Tools to Improve AI Image Provenance

Since 2024, OpenAI has added C2PA Content Credentials to images from DALL·E 3, ImageGen and Sora, recording their provenance through metadata and cryptographic signatures. That information, however, can be removed through format conversion, resizing or screenshots. OpenAI's adoption of Google DeepMind's SynthID adds an invisible signal embedded in pixels, creating a two-layer provenance system.

OpenAI announced on May 19, 2026, that images generated by ChatGPT, Codex and the OpenAI API would include SynthID, and previewed a public verification tool. Users can upload a single PNG, JPG or WEBP file for separate C2PA and SynthID checks. The tool currently recognizes only OpenAI content. Neither company disclosed the value of the partnership, and the absence of a detected signal does not mean an image was not AI-generated.