What it means
Content Credentials, built on the C2PA specification, attach a signed manifest to a file recording its origin and edit history: what device or model produced it, what changes were applied, by whom.
The strategic insight is the direction of proof. Detecting synthetic media is a losing race — detectors and generators improve against each other indefinitely. Provenance inverts it: instead of catching every fake, establish what is verifiably authentic, and treat unsigned material as simply unverified.
Adoption spans camera manufacturers, editing software and several AI model providers, though it is far from universal. The known weakness is fragility — metadata is commonly stripped by ordinary handling, and screenshotting or re-encoding removes it entirely. Absence of credentials therefore proves nothing.
Why it matters
This is the most credible structural answer to synthetic media, and it is why 'just build a detector' is the wrong ask. It only works at scale, though: credentials are useful in proportion to how many creators sign and how many platforms display them.
In practice
Treat credentials as positive evidence only. Present material is worth checking; absent material is unverified rather than fake — and given how routinely metadata is stripped, that distinction matters.
Where this shows up
Tools and models in our catalog.
