OpenAI's C2PA and SynthID watermark make AI images traceable

OpenAI announced two complementary measures to help verify whether an image was generated by its AI models. The first is adoption of the C2PA (Coalition for Content Provenance and Authenticity) standard, a non-profit initiative founded in 2021 that embeds provenance metadata directly into image files. C2PA signals are accessible but can be stripped by bad actors. The second is an invisible watermark called SynthID, developed by Google, which is designed to survive screenshots, resizing, and other digital manipulation. OpenAI notes that together, metadata and watermarking make provenance more resilient than either layer alone, addressing each other’s weaknesses.

OpenAI is also previewing a public verification tool that checks for both C2PA and SynthID signals, allowing users to test whether an image originated from an OpenAI product. Initially, the tool only supports images from OpenAI’s own generators (e.g., DALL·E), but the company aims to expand coverage to other tools over time. The move positions OpenAI as a proactive player in combating AI-generated misinformation, though industry-wide adoption of C2PA remains inconsistent—Google already uses it across some products, but many less reputable tools lack such protections.