OpenAI announced two measures to help verify whether an image came from one of its models: joining the C2PA standard (Coalition for Content Provenance and Authenticity) and adding Google's SynthID watermarking to its products.
C2PA is an open technical standard that encodes provenance metadata directly into an image file, recording where content came from and how it was created. A compatible reader can check a C2PA-tagged image and surface whether it was photographed, edited, or AI-generated. Images from ChatGPT's image tools and DALL-E will now carry this embedded record.
SynthID is Google DeepMind's invisible watermarking system. It imprints a pattern into the image pixels themselves - one that survives common edits like cropping, resizing, and compression - detectable by a dedicated scanner but invisible to the human eye.
The two technologies cover different vulnerabilities. C2PA metadata lives in the file's header and can be stripped by resaving the image. SynthID is woven into the pixels and can't be removed without visibly degrading the image. Using both together closes that gap.
Neither measure stops determined bad actors - there are workarounds for both. But for the far more common case - an editor checking a submitted photo, a platform moderating uploaded content, or someone who wants a quick verification signal - these tools create a usable paper trail. OpenAI handles a large share of AI image generation across its products, so its adoption gives C2PA meaningful momentum toward the industry-wide critical mass these standards need to actually function.