Question: when I upload an OpenAI-generated PNG with valid C2PA/Content Credentials into ChatGPT, the file available inside the chat appears to be stripped down to only IHDR/IDAT/IEND chunks. The same original file verifies fine on openai.com/research/verify with both SynthID and Content Credentials detected.
Is ChatGPT’s upload pipeline intentionally sanitizing/removing C2PA chunks, and is openai.com/verify the intended/only supported way to verify provenance? Or is there a way to preserve/read the original uploaded file metadata inside a chat?
The image in question is:
what ChatGPT says he sees:
and what the openai.com/research/verify website sees:
