Education#content credentials#c2pa#provenance

What Are Content Credentials? A Plain-English Guide to C2PA

Content Credentials are signed C2PA records of how an image was made. What they contain, who adds them, where you see them, and their limits in 2026.

BA
BlankAI Team·7 min read · Published · Updated

The people who build BlankAI

In This Article

  1. 01What a manifest records
  2. 02Where you see them
  3. 03Who adds them in 2026
  4. 04Durable Content Credentials
  5. 05Their limits
  6. 06Should you keep them?

Content Credentials are a way to attach a tamper-evident history to a file. They are built on C2PA, an open standard from the Coalition for Content Provenance and Authenticity, whose members include Adobe, Google, Microsoft, OpenAI, Sony and TikTok. When an app or camera supports it, it writes a signed manifest into the file that says what made it and what happened to it since.

What a manifest records

  • The claim generator: the app, model or device that signed the manifest.
  • Actions such as created, edited, cropped or converted, with an IPTC digital source type when AI was involved.
  • Ingredients: earlier versions of the file and their own manifests, so a chain of edits can be followed.
  • A digital signature and certificate chain, which lets a verifier check that nothing was changed after signing.

The current specification is version 2.4, dated April 2026. The C2PA now runs a conformance program and an official trust list of certificate authorities, replacing the interim list it froze on 1 January 2026.

Where you see them

Most of the time you do not see them at all. LinkedIn is the main exception: it shows a CR icon on images and videos with credentials, and clicking it opens the details. You can inspect any file at contentcredentials.org/verify, with Adobe's Content Authenticity app, or with BlankAI's EXIF Viewer. Google added C2PA checks to the Gemini app in May 2026 and has said Search and Chrome will follow.

Who adds them in 2026

OpenAI signs every ChatGPT, Codex and API image. Google says all media made in the Gemini app carries them. Adobe adds them to Firefly images and to Photoshop exports that used generative features. Microsoft adds them in Copilot, Designer and Bing Image Creator. Black Forest Labs signs FLUX API output, TikTok signs content made with its AI tools, and Anthropic has signed image files created by Claude since August 2026. On the camera side, recent Sony, Leica, Canon and Fujifilm bodies and the Pixel 10 can sign photos at capture.

Durable Content Credentials

Because a manifest is file data, it is lost whenever a file is re-encoded, screenshotted or uploaded to a service that strips metadata. Durable Content Credentials address this by pairing the manifest with an invisible watermark or a fingerprint, so a verifier can look the manifest up again from a cloud copy. Removing the manifest from a file does not remove that watermark.

Their limits

  • Absence proves nothing: most images online have no credentials, and any re-encode removes them.
  • Presence is evidence, not proof: in August 2026 a researcher forged valid camera credentials on a rooted Pixel 10, and Google classed the issue as won't fix.
  • Credentials describe the file's history as recorded by the signing apps. They say nothing about whether the scene is true.

Should you keep them?

Keep them when they work for you: a photojournalist's capture record, a portfolio that shows your process, a client who asks for provenance. Consider removing them from copies when they expose things you did not intend, such as a tool chain, account names or a minor AI retouch that triggers a label. If the image is AI-generated and presented as real, keep the label.

BlankAI shows whether a file has a C2PA manifest and which organizations it mentions, without uploading the file.

Check a file
#content credentials#c2pa#provenance#cr icon

Share this article

About the authors
BA

BlankAI Team

The people who build BlankAI

BlankAI builds browser-based tools for inspecting and cleaning image metadata. These guides are written by the same team, checked against vendor documentation, standards and independent tests, and updated when those change. Sources are listed at the end of each article. Corrections: [email protected].