Content provenance
Information attached to or embedded in media that records where it came from, including whether AI generated it, such as C2PA Content Credentials or invisible watermarks.
Why it matters. Some platforms and regulations expect it; metadata can be stripped, watermarks are more robust.
How it works
Content provenance is information that records a piece of media's origin and edit history, including whether AI generated or altered it. The two main approaches are signed metadata attached to the file and watermarks embedded in the content itself.
C2PA, from the Coalition for Content Provenance and Authenticity, is an open technical standard for cryptographically signed provenance metadata, shown to people as Content Credentials. Metadata is easily lost through screenshots or re-encoding, so some makers also embed invisible watermarks, such as Google's SynthID. Many image and video providers attach one or both by default.
If your product generates media, check what the model adds and whether your pipeline keeps it intact. Platform policies and some regulations increasingly expect AI-generated content to be disclosed.
Worked example
Illustrative: an app generates a C2PA-signed image, then resizes and re-encodes it with a tool that drops metadata. The published file no longer carries its Content Credentials, though a watermark in the pixels, if the model added one, may still be detectable.
Common mistakes
- Running generated media through processing that strips C2PA metadata without noticing.
- Treating missing provenance data as proof that content is human-made.
- Assuming every model adds the same provenance signals; check each maker's docs.
Across the catalog today
See it on a real card: Safeguards on Veo 3.1.