Content Credentials (C2PA): How the Photo Label Works
Short answer
Content Credentials are signed records, defined by the C2PA standard, that travel with an image or video and state which device or app created it and what was changed. A valid credential proves where a file came from. A missing one proves nothing, because most platforms still strip metadata on upload.
Open almost any news photo on a social network and look for a small “CR” icon. You will rarely find one. The technology to label images with their origin has existed for years, it is backed by Adobe, Google, Microsoft, OpenAI, Meta, Sony, the BBC and others, and phones now sign photos by default. Yet the label usually disappears somewhere between the camera and your screen.
That mismatch explains most of the confusion about Content Credentials. The system is good at proving that a file is authentic when the credential survives. It is useless for proving that a file is fake when the credential is missing.
What Content Credentials are
C2PA stands for the Coalition for Content Provenance and Authenticity. It publishes an open technical standard, currently version 2.3, for attaching a signed history to digital media. Content Credentials is the consumer-facing name for the result, which the coalition describes as working “like a nutrition label for digital content” (C2PA).
Inside the file sits a manifest. It contains assertions such as the device or software that created the file, the time, the edits applied, and whether a generative AI tool was involved. The manifest is signed with a certificate, and it is bound to the exact content with a cryptographic hash. Change a single pixel without updating the manifest and validation fails.
- CaptureA camera or phone app signs the original: device class, time, sometimes location.
- EditAn editor that supports C2PA adds a new manifest listing the changes and links it to the earlier one.
- GenerateAI tools that support the standard record that content was created or altered by a model.
- PublishA platform that keeps metadata shows the CR icon. Many platforms remove it here.
- VerifyAnyone can upload the file to a verify tool and read the full chain.
Which devices and apps sign today
The steering committee includes Adobe, Amazon, the BBC, Google, Meta, Microsoft, OpenAI, Publicis, Sony, TikTok and Truepic. Membership does not mean every product signs content, so check the specific tool.
Phones are where the volume is. Google says the Pixel 10 is the first phone line with Content Credentials on every photo taken in the Pixel Camera app, and that the app reached Assurance Level 2, “the highest security rating currently defined by the C2PA Conformance Program”. Google Photos can display credentials for JPEGs and attaches them to images edited with its AI tools (Google).
Among cameras, Leica’s M11-P from October 2023 was the first model sold with Content Credentials built in. Several Sony and Nikon bodies followed through firmware or cloud services. The rollout has not been smooth: Nikon suspended its service for the Z6III in 2025 after a security problem was reported and revoked the affected certificates. That episode is a useful reminder that the trust lives in the signing keys, not in the logo.
Why the label is usually missing
Three things strip credentials in everyday use. None of them is malicious.
- Platforms. Many social networks and messengers re-encode uploads and remove metadata to save space and protect privacy. The credential goes with it.
- Screenshots. A screenshot is a new file created by your phone’s operating system. It carries none of the original’s history.
- Old software. Editors that don’t understand C2PA may save the image without the manifest, or with a broken one.
The standard has an answer called soft binding. An invisible watermark or a content fingerprint is stored with the manifest in a repository, so a verifier that finds the watermark can look up the credential even after the metadata is gone. Adobe uses this approach for its own Content Credentials. The trade-off: you are now trusting a lookup that says “something like this was signed once”, not a signature on these exact bytes.
| Criterion | Embedded manifest | Watermark or fingerprint lookup |
|---|---|---|
| Survives a screenshot | partly | |
| Survives platform re-encoding | partly | |
| Proves these exact pixels were signed | ||
| Needs an online repository | ||
| Works offline |
Who benefits most right now
For ordinary users scrolling a feed, Content Credentials are still mostly invisible. The people who gain most today are those who receive original files: picture desks, fact-checkers, insurers assessing damage photos, lawyers handling evidence and HR teams checking documents sent in by applicants. They can insist on originals and reject anything that arrives as a screenshot.
Photographers gain too. Signing your work at capture gives you a record that predates any copy, which helps in disputes over authorship. It does not stop anyone from stealing the picture, but it makes the theft easier to show.
How to check a file yourself
You need the original file, not a screenshot. If someone sent you a picture through a messenger, ask for the file as a document or a download link.
- Open contentcredentials.org/verify and upload the file. It accepts JPEG, PNG, WebP, HEIC, AVIF, TIFF and several video and audio formats.
- Read the summary: who signed it, with which device or app, and whether AI tools were used.
- Open the history to see each step. Check that the dates make sense.
- If the tool says “no Content Credentials found”, that is a neutral result. Move on to other checks.
On the command line, ExifTool by Phil Harvey reads C2PA data along with EXIF, IPTC and XMP (ExifTool). Running exiftool -a -G photo.jpg lists every tag grouped by type, which helps when a verify site rejects a file.
What a credential does not tell you
Valid credentials say the file came from a particular device or app and was not altered afterwards. It does not say the scene is true. Real cameras can photograph staged events, and phones can photograph a screen showing an AI image. The credential will then honestly report “taken with this phone”. Nothing in the standard checks what was in front of the lens.
It also tells you nothing about intent. News agencies and propaganda outlets can both sign their photos. What you gain is accountability: you know who stands behind the file, and you can judge that source.
For AI images specifically, credentials complement watermarks. Google’s SynthID, for example, is checked separately in the Gemini app. Our guide on how to tell if an image is AI generated combines both checks with visual inspection.
Where this is heading in the EU
From 2 August 2026, Article 50(2) of the EU AI Act requires providers of generative AI to mark output in a machine-readable way, where technically feasible. Systems already on the market before that date have until 2 December 2026. Signed metadata such as C2PA is one way to meet that duty, usually combined with watermarks because metadata alone is too easy to strip. The details, including fines, are in our explainer on EU AI Act Article 50.
If you publish images at scale, the practical step is to keep credentials intact in your own pipeline: sign on capture or export, avoid tools that strip metadata, and show the CR icon where your CMS supports it. Teams that need help wiring provenance into an existing content system can talk to OnlineBuilders. For one-off checks, start with the verify tool and then a reverse image search.
Questions people ask
Is C2PA the same as a watermark?
No. C2PA is signed metadata attached to the file. A watermark is a signal hidden in the pixels or audio. The standard can link the two, so a watermark can point to a stored copy of the credential after the metadata has been removed.
Can I add Content Credentials to my own photos?
Yes. Some cameras and phones sign at capture, and editing apps such as Adobe Photoshop can attach credentials when you export. Check the export settings of your tool.
Does a Content Credential show my name or location?
Only if you choose to include it. Google says its Pixel implementation is designed so that images cannot be linked to each other or to the person who took them. Other tools let you add your name as the creator.
Can Content Credentials be faked?
The signature itself is hard to forge, but a dishonest device or app can sign false information, and a stolen signing key can sign anything. That is why the C2PA runs a conformance program that rates how well implementations protect their keys.
.jpg){kind=link}