Provenance (Content Credentials)

When you see a photo online, you usually don’t know where it came from. It could have been captured with a camera, edited on a computer, or generated entirely by a computer program. Provenance is the technical term for the origin and history of such a file. Content Credentials are a unified format in which this history is stored directly within the file itself. You can think of it like a digital label stuck to the file, recording: captured here, modified there, then published. Anyone who later views the file can read this label and check whether it was tampered with along the way.

What proof of origin achieves against fakes

Programs today generate images and videos that are barely distinguishable from real recordings. Such fabricated recordings are called deepfakes. They show up in election campaigns, in fake advertisements featuring well-known faces, and in scams. The obvious reflex is to build software that detects fakes. That only works moderately well, because the generation programs keep getting better.

Provenance turns the task around. Instead of exposing fakes, it proves the authenticity of genuine content. A news agency can thus demonstrate that its war photo actually came from a specific camera. An image with no proof of origin at all isn’t automatically false, but it is simply unverified. This shifts the burden of proof.

Lawmakers are also interested in this. The European Union’s AI Act requires that artificially generated content be labeled in a machine-readable way. Content Credentials are one way to fulfill this obligation in practice. For companies like Adobe, Microsoft, or OpenAI, this is a reason to build in the standard.

From camera sensor to verification page

It starts with the device or program that creates the content. It writes information into the file: timestamp, device used, and if applicable, a note that AI was involved. This information is then digitally signed. A digital signature is a kind of seal that only the issuer can create and that anyone can verify.

Every edit adds a new entry, along with a new signature. This creates a chain. If someone alters the image without a matching entry, the seal no longer matches the content, and verification fails. The technical rules for this are set by a coalition of companies, the C2PA. Its specification is open, and anyone may implement it.

A common misconception: Content Credentials don’t prevent forgery. They only prevent a forgery from passing unnoticed as authenticated. And they’re easy to remove, for instance by taking a screenshot. That’s why some providers additionally embed a watermark within the image itself — an invisible pattern that survives such workarounds.

A small icon in cameras, photo editors, and feeds

Cameras from Leica, Nikon, Sony, and Canon can already sign photos the moment they’re taken. In Photoshop, the edit history can be attached as a Content Credential. Image generators like DALL·E or Firefly mark their outputs as AI-generated by default.

This becomes visible through a small icon, usually a clipped square with the letters cr. LinkedIn, TikTok, and Google display such information for some content. On the website contentcredentials.org, you can upload a file and view its stored history.

In business news, you’ll mostly encounter this term in connection with regulation, copyright, or trust in media. Don’t confuse provenance with the watermark: the watermark is embedded in the image content itself, while provenance data sits alongside it as supplementary information. In practice, both methods complement each other.