OpenAI Adds SynthID Watermarking Alongside C2PA to All Generated Images
OpenAI has committed to applying both C2PA content credentials and Google's SynthID invisible watermarking to images produced by its models across ChatGPT, Codex, and the API. The move brings OpenAI into alignment with the two most widely adopted provenance standards currently in circulation, and marks a significant expansion in the number of images that will carry detectable AI-origin signals.
The two systems address different failure modes. C2PA embeds structured metadata describing how and where content was created or modified - useful when that metadata survives distribution. SynthID, developed by Google DeepMind, encodes information directly into the pixel or audio data itself, meaning a watermark can persist even when metadata is stripped during screenshot, recompression, or social media upload. OpenAI described the combination as a "multi-layered approach," noting that the systems reinforce rather than duplicate each other.
OpenAI also announced a verification tool to let users check whether an image carries these markers - a practical step toward making provenance checks accessible outside of specialist tools or platform-level integrations.
The timing coincides with Google's own SynthID expansion at I/O 2026, where the company announced that Chrome and Google Search features including Lens and AI Mode would begin surfacing SynthID and C2PA verification results. Nvidia is also among the companies newly adopting SynthID. The collective adoption suggests these two standards are converging toward something close to an industry baseline for AI image labeling, though widespread consumer awareness of what the markers mean remains a separate challenge.
