PricingLog in

Why metadata gets stripped.

Content Credentials travel as metadata inside the image file, and metadata is lost when a platform or tool re-encodes the image without copying it. A watermark or fingerprint stored as a soft binding is how the C2PA specification lets a lost manifest be found again.

How it is lost

OpenAI, when it added C2PA metadata to generated images, noted that metadata can be removed and that some social media platforms automatically remove metadata from uploaded images (reported by The Decoder, 6 February 2024). Any step that writes a new file without carrying the manifest across has the same effect.

What can survive, by design

  • Hard binding — cryptographic hashes that uniquely identify the asset or a portion of it. It proves the file is the one signed, so it does not survive re-encoding; a changed file no longer matches.
  • Fingerprint — properties computable from the content that identify it or near-duplicates, so a copy can be matched without any embedded data.
  • Invisible watermark — information incorporated into the content in a substantially human-imperceptible way, usable to identify the asset.

Fingerprints and watermarks are soft bindings. With a manifest stored in a repository, they make a credential durable: findable even after the embedded copy is gone.

Limits

  • A soft binding helps only if a manifest copy was stored somewhere it can be looked up.
  • No mark is guaranteed to survive every edit or re-encoding, and none is a guarantee of detection.
  • A file with no credentials is not proof of anything. Try the Content Credentials checker to see what a file still carries.

Proofcan writes all three layers for the images it marks: a C2PA manifest, an invisible watermark carrying a lookup ID, and a registry record.

Questions

Do social platforms remove image metadata?
Some do. Reporting on OpenAI’s C2PA rollout quotes OpenAI noting that metadata can be removed and that some social media platforms automatically remove metadata from uploaded images.
What is a soft binding?
In the C2PA Specification, a content identifier that is either not statistically unique, such as a fingerprint, or embedded as an invisible watermark in the identified digital content.
What is a Durable Content Credential?
A Content Credential for which there exist one or more soft bindings that enable its discovery in a manifest repository.
Can a removed manifest be recovered?
The specification describes the case: if a manifest is removed from an asset but a copy remains in a provenance store elsewhere, the manifest and asset may be matched using available soft bindings. That requires the copy to have been stored somewhere to begin with.

Sources

Page reviewed . Proofcan is a technical tool, not legal advice, and does not guarantee compliance.