OpenAI content provenance: C2PA + SynthID layering

Start here

In everyday words

OpenAI is adding stronger “where did this come from?” signals to its images, using both metadata and watermarking, and says it plans to offer a way for the public to verify OpenAI image origin.

Need a meaning?

Quick Sip

What you need to know

Who is affected: journalists, platform integrity teams, everyday social media users
What changed: On May 19, 2026, OpenAI published an update on content provenance, saying it is now a C2PA Conforming Generator Product, will add SynthID watermarking to images generated through ChatGPT, Codex, and the OpenAI API, and is previewing a public verification tool to check whether an image came from OpenAI.
Why it matters: Content provenance only works at scale if signals survive reposting and remain machine-readable across platforms. By combining signed metadata (C2PA) with an embedded watermark (SynthID), OpenAI is betting on a layered approach that makes AI-generated images easier to identify even after common transformations.
What to watch next: Watch whether major platforms preserve and display C2PA metadata end-to-end, how well SynthID detection holds up under editing and recompression, and what the public verification tool can (and cannot) reliably determine.

Four useful details

OpenAI says it is now a C2PA conforming generator, so platforms can more reliably read and preserve provenance metadata it attaches to content.
It says it will add Google DeepMind’s SynthID watermarking to OpenAI-generated images as a durability layer when metadata is stripped or lost.
OpenAI says SynthID watermarking will start with images generated through ChatGPT, Codex, and the OpenAI API.
The company also previewed a public verification tool intended to help check whether images came from OpenAI.

OpenAI · Official AnnouncementOffering Zero Data Retention for frontier models

Adds source-backed context on ai safety from OpenAI.