C2PA Content Credentials

Cryptographically signed metadata attached to a file, recording what produced it and how it was edited. It is an open standard, and OpenAI attaches it to images from its tools. It sits alongside the content rather than being hidden inside it.

How it works

C2PA attaches a manifest to a file: a record of what produced the content and what was done to it, hashed and signed with a certificate chain that a verifier can check. It sits in the file's metadata rather than in the pixels or the words. Verification is a cryptographic operation with a binary outcome, which is why the approach has no false positives, unlike a statistical watermark that must weigh evidence and report a probability. Its weakness follows equally directly: metadata is trivially stripped, deliberately or as a side effect of any tool that re-encodes a file, and re-encoding is exactly what most upload pipelines do. Absence of a credential therefore proves nothing at all. OpenAI attaches C2PA to images from its tools and uses SynthID for audio; none of it applies to the text you copy out of a chat, which carries no metadata at all.

What it is legitimately for

Provenance. Signed metadata is checkable and, unlike a statistical watermark, has no false positives: either the signature verifies or it does not.

What it breaks

• Nothing in your text. This is not a text technique at all.
• The metadata is easily stripped, deliberately or by any tool that re-encodes a file, so its absence proves nothing.

How to find it yourself

Inspect the file's metadata with a C2PA-aware viewer. Nothing in the text of a chat reply carries it.

What Clipboard Sanitizer does

Nothing, and nothing is needed. Clipboard Sanitizer works on text on your clipboard; this is metadata on image and audio files.