OpenAI will watermark ChatGPT images with Google's SynthID and launch a provenance checker
Images from ChatGPT, the OpenAI API, and OpenAI Codex will carry SynthID and C2PA signals, with a checker to verify provenance and new participation in C2PA conformance.
By Ryan Merket ·
Why it matters
If durable, layered provenance becomes standard, platforms and newsrooms get a workable path to label AI media and triage deepfakes. OpenAI moving first pressures other model providers to adopt compatible signals, but limits remain: metadata is fragile, watermarks can be evaded, and early scope only covers OpenAI images.
OpenAI is adding Google's SynthID watermarks to images generated by ChatGPT, the OpenAI API, and OpenAI Codex, while previewing a public portal to verify provenance, according to The Verge.
The company is pairing SynthID with C2PA content credentials as a multi-layer approach. OpenAI said the layers complement each other by persisting through edits while also carrying richer context, per the report. The goal is to make AI labeling more resilient when metadata gets stripped or screenshots blur provenance.
OpenAI is also previewing a verification portal that checks for C2PA and SynthID signals and flags if an image was generated with ChatGPT, the OpenAI API, or OpenAI Codex. It starts with OpenAI-generated images and is limited to one upload at a time, with plans to support more verification systems and content types over time. The company cautions detection is not foolproof and avoids definitive calls when signals are missing, The Verge notes.
Separately, OpenAI has joined the C2PA Conformance Program, which validates that products correctly implement Content Credentials. OpenAI has embedded C2PA in image and video for some time, but those signals are often lost when content is reposted or platforms strip metadata, a gap watermarking aims to narrow.