OpenAI adds SynthID watermarking and a public content-verification tool
The invisible watermark, developed by a rival lab, is paired with C2PA metadata credentials because OpenAI said neither signal survives image manipulation on its own.
- Culture & impact
- Minor
OpenAI said it would embed Google DeepMind’s SynthID invisible watermark into images generated by ChatGPT, Codex and the OpenAI API, and joined the Coalition for Content Provenance and Authenticity (C2PA) as a “conforming generator,” attaching Content Credentials metadata to the same outputs.
The two signals were presented as complementary rather than redundant: C2PA metadata carries richer information about an image’s origin but can be stripped by re-saving or editing a file, while SynthID’s watermark is designed to survive common manipulations such as cropping and compression but carries no contextual detail on its own. OpenAI also previewed a public tool letting anyone upload an image to check for supported provenance signals, and said it would offer API access so other organisations could build verification into their own products.
The move made OpenAI and Google DeepMind, competitors on model releases, aligned on the same provenance approach for image outputs — pairing an open technical standard with a watermark originally built by a rival lab. It followed a period of rising deepfake and AI-image-authenticity concerns and preceded OpenAI’s own election-safeguards announcement the same month, which cited the same watermarking work as one of its defences against synthetic election imagery.