What Are Content Credentials (C2PA)? A Plain-English Guide

You may have seen a small "CR" icon on an image, or read that ChatGPT and Google now "label" their AI images. Both are Content Credentials, built on an open standard called C2PA. This guide explains what they are, in plain English.

The one-sentence version

Content Credentials are a signed record stored inside a file that says how the image (or video or audio) was made and changed, such as which camera or app created it and whether AI was used, protected so that tampering can be detected.

What does C2PA stand for?

C2PA is the Coalition for Content Provenance and Authenticity, formed in 2021 by companies including Adobe, Arm, the BBC, Intel and Microsoft. It publishes the open technical standard. "Content Credentials" is the friendly name for records made with that standard. The Content Authenticity Initiative, a related Adobe-led community, promotes their use.

What's inside a Content Credential?

The record (called a manifest) has a few parts:

• Assertions: statements about the content, such as "created with this app", "cropped", "background generated with AI", or the files it was made from.
• Digital source type: a standard label for how the content was made. trainedAlgorithmicMedia means AI-generated; compositeWithTrainedAlgorithmicMedia means partly AI.
• A link to the exact content: a cryptographic fingerprint (hash) of the image, so the record can't simply be copied onto a different picture.
• A signature from whoever issued the record (an app maker, camera maker or AI company), with the date it was signed.

Each time a supporting app edits the image, it can add a new layer to the record, so you get a history: created here, edited there, exported as PNG. The earlier layers are kept as "ingredients".

Who adds Content Credentials?

• AI image tools: OpenAI's image tools, Adobe Firefly and Google's newer image models, among others.
• Editing apps: Adobe Photoshop and Lightroom, when the option is on.
• Cameras and phones: Google's Pixel 10 signs photos in its camera app, and some cameras from makers such as Leica and Sony can sign photos at capture.
• Platforms: LinkedIn reads them and shows a "CR" label; see what LinkedIn's CR label means.

What Content Credentials prove, and what they don't

They can show:

• which tool or device made or edited the file, and whether AI was involved, according to that tool;
• who signed the record and when;
• that the record and the image haven't been changed since signing (if the signature checks out).

They can't show:

• That the picture is true. A real photo can still be staged or captioned misleadingly.
• That an image without them is real. Most images have none, because many tools don't add them and screenshots and social apps often strip them.
• That the signer is trustworthy. A valid signature proves who signed, not that they told the truth. Verifiers check signers against a list of trusted issuers for this reason.

Content Credentials vs. watermarks

Content Credentials live in the file's data, so anyone can read them with an open tool, but they can be removed. Invisible watermarks such as Google's SynthID are woven into the pixels and survive screenshots better, but only the company that made them can check them. The two work best together; see how to check images from ChatGPT and Gemini.

How to read the Content Credentials in an image

1. Open the OriginLens Image Analyzer and add the image. It runs in your browser; nothing is uploaded.
2. The File labels card tells you if the record declares AI generation, and whether the signature checks out.
3. The Provenance tab shows the full record: the tool, the signer, the date, every digital source type and any validation issues.

OriginLens reads every layer of the history, so an AI step recorded one edit back (for example, before a watermark or format conversion) is still found. For an official check of whether the signer is on the trusted list, you can also use the Content Authenticity Initiative's Verify site.

Privacy: what Content Credentials reveal about you

A record can include your name, your device, the apps you used and your edit history, and anyone with the file can read it. If that matters, check what's in your file before you share it, and see our guide on removing metadata from photos.

Frequently asked questions

Is C2PA the same as Content Credentials?

Almost. C2PA is the standard and the group that writes it; Content Credentials are the records made with it. People use the two terms interchangeably.

Do all AI images have Content Credentials?

No. Many AI tools still don't add them, and they are easy to lose when an image is screenshotted or re-uploaded. A missing record proves nothing either way.

Can Content Credentials be faked?

Changing an existing record breaks its signature, which a verifier will show. But someone can create a new record with false claims and sign it themselves, which is why who signed it matters as much as what it says.