AI images
Content Credentials (C2PA) explained: how to verify where an image came from
Content Credentials are a signed record, defined by the C2PA standard, that says which app, AI model or camera made a file and how it was edited. Check an image for free with the uncrop.si AI image checker; a video, audio file or PDF with the Content Authenticity Initiative's Verify site; and images on web pages with Adobe's free browser extension. A trusted record is strong evidence but not proof, and screenshots and many apps remove credentials, so a missing record proves nothing.
Content Credentials are a signed history attached to an image. They record which app, AI model or camera made it, what was done to it afterward, and who vouches for that record. They are built on an open standard from the C2PA, and most of the big AI image tools now add them. This guide explains what is inside them, what “trusted” and “valid” mean, who adds them, and how to check an image yourself.
What is C2PA?
C2PA is an open standard, written by the Coalition for Content Provenance and Authenticity, for recording where a photo, video or audio file came from. A record made under it is called a Content Credential: a signed history of the app, AI model or camera that made the file and the edits that followed. In C2PA terms, that record is called a manifest, and each statement in it is an assertion (C2PA Explainer).
The Coalition for Content Provenance and Authenticity (C2PA) was announced in February 2021 by Adobe, Arm, BBC, Intel, Microsoft and Truepic to fight misleading content with an open provenance standard (C2PA). Its steering committee now includes Adobe, Amazon, BBC, Google, Meta, Microsoft, OpenAI, Publicis Groupe, Sony, TikTok and Truepic (C2PA). The C2PA writes the standard; the Content Authenticity Initiative, which Adobe founded in 2019, builds the open-source tools and calls the result Content Credentials (Content Authenticity Initiative).
The record lives inside the image file, signed with a certificate the way a secure website is. If anyone changes the signed parts of the image or the record afterward, the signature no longer matches.
What a Content Credential records
Here is what we found in a ChatGPT image made in October 2026, read with the uncrop.si AI image checker:
| Field | What it said |
|---|---|
| Signed by | OpenAI OpCo, LLC, with a certificate from a C2PA certificate authority |
| Written by | OpenAI Media Service API |
| How it was made | Created with gpt-image, OpenAI’s image model, digital source type “trained algorithmic media”, which means generative AI |
| Later steps | Converted, and an invisible watermark added to the pixels |
| When | The signing time, from OpenAI’s time-stamping service |
It did not contain the prompt, and it did not name ChatGPT: this record names OpenAI’s model and media service, so it shows that OpenAI made the picture, not which OpenAI app was used. Other records can list more: a camera model, each edit made in an app such as Photoshop, and the earlier files (“ingredients”) an image was built from.
A credential can also say who made the work. Adobe Content Authenticity, a free web app (Adobe), lets creators attach their name and social media accounts to their own images, plus a preference on whether the work may be used to train generative AI (Adobe). The uncrop.si checker does not show a credential’s creator name or that preference, and it does not add Content Credentials to files: it only reads and verifies them.
How to check or verify Content Credentials for free
- Use the original file. A screenshot, or a copy downloaded from most social apps, usually has no credentials left.
- For an image, drop it into the uncrop.si AI image checker. It reads the record and verifies the signature on your device, against the C2PA trust lists, without uploading the image. It tells you whether the record is trusted, valid or invalid (explained below), along with what it says about AI.
- For a video, audio file or PDF, or for a second opinion, upload it to the Content Authenticity Initiative’s free Verify site at verify.contentauthenticity.org, which contentcredentials.org/verify also leads to. It accepts AVI, AVIF, DNG, HEIC, HEIF, JPEG, M4A, MOV, MP3, MP4, PDF, PNG, SVG, TIFF, WAV and WebP files, and shows the full history, including earlier versions of the image when they are recorded.
- To check images and videos on a web page without downloading them, add Adobe’s free Content Authenticity extension to Chrome or another Chromium-based browser (Adobe). A “cr” pin appears on content that carries credentials: select it for a quick summary, then select Inspect to see them in more detail in Adobe’s Inspect tool (Adobe). The extension shows only the most recent credentials in a file (Adobe).
Trusted, valid and invalid
The C2PA specification grades a record in steps, and every trusted record is also valid (C2PA Technical Specification):
- Valid: the record is intact, the file has not changed since signing, and the signature checks out with a certificate that was valid and not revoked at the time.
- Trusted: valid, and the signer’s certificate also leads back to a trust list. Checkers must include the official C2PA Trust List.
- Invalid: something does not match, usually because the image was changed after it was signed.
The C2PA’s July 2026 deployment guidance puts it simply: valid means intact, while trusted means intact and signed by someone on the C2PA Trust List (C2PA). The official list comes from the C2PA conformance program, which launched in 2025. An older interim list was frozen on 1 January 2026, and content signed under it while its certificate was valid still counts (C2PA). The uncrop.si checker uses both lists.
That is also why a genuine camera photo can come out as valid rather than trusted. Sony, for example, provides a default signing certificate free of charge, but says it “does not guarantee the validity of the C2PA signature” and recommends a certificate equivalent to one on a trust list instead (Sony). A photo signed with a certificate that is on no trust list may be perfectly real, but a checker cannot confirm who signed it.
Who adds Content Credentials
| Source | What gets Content Credentials |
|---|---|
| OpenAI | Images from ChatGPT and the OpenAI API, since early 2024 (OpenAI) |
| Images from Nano Banana Pro in the Gemini app, Vertex AI and Google Ads, since November 2025 (Google) | |
| Adobe Firefly | Files where every pixel comes from Firefly, such as Text to Image (Adobe) |
| Microsoft | Images generated or edited with AI features in Designer (Microsoft). In 2023 Microsoft also said it would mark images from Bing Image Creator (IPTC) |
| Samsung | The Galaxy S25 series supports Content Credentials, which Samsung adopted for content created and edited with generative AI (Samsung) |
| Google Pixel 10 | Every photo from the Pixel Camera app, and AI-edited images in Google Photos (Google) |
Cameras can sign photos too. Leica says its M11-P, announced in October 2023, was the first camera to attach Content Credentials at the moment of capture, once the “Leica Content Credentials” menu setting is turned on (Leica). Sony’s α1, α1 II, α9 III, α7 IV, α7 V, α7 S III, α7R V, FX3, FX30 and PXW-Z300 can add C2PA signatures to photos, video or both, with a camera license from Sony (Sony). For any other camera, check the maker’s documentation for your model.
Where you will see them
The official Content Credentials icon is a small “cr” pin that the C2PA introduced in October 2023 (C2PA). Some sites, such as LinkedIn and Behance, show it on content that carries credentials, while others, such as YouTube, display credentials differently (Adobe). The pin means the file carries C2PA data, and hovering over it or selecting it opens a summary of who made the file and how. If the record fails its checks, the C2PA’s guidance lets the site add a second status mark to the pin (C2PA UX guidance). A missing pin proves nothing either way, because many sites remove credentials on upload.
- LinkedIn shows the details when you click the icon: the app or device, the issuer, the signing date and any AI use the signer declared. LinkedIn notes it cannot yet label all AI content (LinkedIn Help).
- YouTube shows “Captured with a camera” under “How this content was made” when a video comes from a capture tool with built-in Content Credentials and was not edited in a way that breaks them (YouTube Help).
- Google Search can show whether an image was captured with a camera or generated with AI, in the “How the image was made or edited” part of “About this image”, when those details are available. Google says this metadata may not be accurate and the feature is available only in some regions (Google Search Help). Google said in September 2024 that this would use C2PA metadata (Google).
- Gemini, at gemini.google.com and in the Gemini app for Android, checks Content Credentials in image, video and audio files from products on the C2PA Conforming Products List, for version 2.2 of the standard and later. It names the company that signed them and summarizes the edit history and any AI use (Gemini Apps Help).
- TikTok says it has labeled more than 3 billion items as AI-generated, using Content Credentials alongside invisible watermarks and its own labeling tools (C2PA).
Credentials kept online
Content Credentials do not have to sit inside the file. The C2PA specification lets an app store the record on a server and put only a link in the file’s XMP metadata, under the key dcterms:provenance (C2PA Technical Specification). The Content Authenticity Initiative’s documentation says only Adobe runs such a cloud service so far (Content Authenticity Initiative).
Readers treat these links differently. Gemini’s help says it does not support remote metadata (Gemini Apps Help). The uncrop.si AI image checker shows the link and where it points, but does not fetch the record, so checking a file never tells a server what you are looking at. In our test with a sample file from the Content Authenticity Initiative, the link pointed to cai-manifests.adobe.com, and the signed record was still there to download in October 2026. When you need to see such a record, open the file in a C2PA reader that is allowed to fetch it.
Why credentials disappear
The C2PA’s own guidance warns that screenshots, and exporting through tools that strip metadata, remove embedded Content Credentials, and that a “no Content Credentials found” result can simply mean they were lost along the way (C2PA). That is why companies such as OpenAI and Google pair them with an invisible watermark, which is built to survive those steps. See what SynthID is and how to check it. To see which metadata a file still holds after it was shared, open it in the image metadata viewer.
Stripped credentials can sometimes be found again. The C2PA standard allows a “soft binding”, an invisible watermark or a fingerprint of the image, that lets a credential be looked up in cloud storage after it was removed from the file (C2PA Explainer). In Adobe Content Authenticity, Search for possible matches can recover credentials stored in Adobe’s Content Credentials cloud, even from a screenshot. If they were only embedded in the file and a site stripped them on upload, they may not be recoverable this way (Adobe). The uncrop.si checker reads only what is in the file and does not search for matches.
Content Credentials, AI labels and SynthID
| Question | Content Credentials | IPTC AI label | SynthID |
|---|---|---|---|
| Where it lives | A signed record in the file | A plain metadata field in the file | An invisible pattern in the pixels |
| Can it be checked by anyone? | Yes, with free tools | Yes | No, only by Google and OpenAI |
| Proves who made it? | Strong evidence, when trusted | No, anyone can write it | That a participating company’s AI made it |
| Survives a screenshot? | No | No | Built to survive many edits |
No single mark covers every case, so check all three. The full routine, including what to do when the file says nothing, is in how to tell if an image is AI-generated.
Check an image for AI labels and watermarks
The free uncrop.si checker reads Content Credentials, AI labels and generator settings on your device, and points you to the SynthID checks from Google and OpenAI.
Questions
What does C2PA stand for?
The Coalition for Content Provenance and Authenticity. Adobe, Arm, BBC, Intel, Microsoft and Truepic founded it in February 2021 to write an open standard for recording where digital content comes from.
Is C2PA the same as Content Credentials?
Not quite. C2PA is the coalition and the open standard it writes. A Content Credential is the plain-language name, used by the C2PA itself and by the Content Authenticity Initiative, for a signed record made under that standard. In everyday use, people often treat the two as the same thing.
Does a valid Content Credential mean the image is real?
No. It shows that the record is intact and the file has not changed since it was signed, not that the scene is true. The C2PA's own explainer says provenance information alone cannot tell you whether content is true, accurate or factual.
What is the difference between valid and trusted Content Credentials?
Valid means the record is intact, the file has not changed since it was signed, and the signature checks out. Trusted also means the signer's certificate leads back to a trust list, such as the official C2PA Trust List.
Do ChatGPT images have Content Credentials?
Yes. OpenAI has added Content Credentials to its generated images since early 2024. ChatGPT images we checked in October 2026 were signed by OpenAI OpCo, LLC and recorded that they were created by generative AI with gpt-image, OpenAI's image model.
What does the cr icon on an image mean?
It is the official Content Credentials icon, which the C2PA introduced in 2023. It tells you the file carries a C2PA record: select it to see a short summary of who made the file, with which tools, and how it was edited.
Do phones add Content Credentials?
Some do. Google says the Pixel 10 adds them to every photo taken with the Pixel Camera app, and Google Photos adds them to images edited with AI. Apple has taken a different route on iPhone 18 Pro and iPhone 18 Pro Max: Apple Reference Image, an opt-in camera mode that Apple contrasts with C2PA-based approaches. Apple says a reference image can be viewed only on an iPhone, iPad or Mac running iOS 27, iPadOS 27, macOS 27 or later, so a C2PA checker such as ours will not show it. You cannot capture reference images in the EU or mainland China.
Can Content Credentials be faked?
It is possible, so treat even a trusted record as strong evidence rather than proof. Changing the signed parts of a file breaks the signature, but the C2PA's own security document notes that anyone who steals a signing key, or is wrongly issued one, can sign any file with a record that passes validation checks, and that a camera fed false data, or pointed at a screen, still makes a technically valid capture. A 2026 security analysis of version 2.2 of the standard found that its exclusion ranges, the parts of a file a record leaves out of its hash, allow undetectable changes, though its authors say some of their suggested fixes were adopted in version 2.3. A researcher also got a rooted Pixel phone to sign an AI image with its camera keys, so that it read as an unedited Pixel Camera photo, and Google closed his report as infeasible to fix. Weaker cases need even more care: a valid record from a signer on no trust list, and an image whose real credentials were removed.
Sources
Steps were checked against official help pages on 22 October 2026. Menus change, so tell us if something moved.
- C2PA founding press release (C2PA)
- C2PA membership and Steering Committee
- How it works (Content Authenticity Initiative)
- C2PA Technical Specification 2.4
- C2PA Explainer 2.4 (C2PA)
- Content Credentials Deployment Guidance, July 2026 (C2PA)
- Conformance program and trust list (C2PA)
- Trust lists (Content Authenticity Initiative documentation)
- Content Credentials Verify (Content Authenticity Initiative)
- Understanding the source of what we see and hear online (OpenAI)
- AI image verification in the Gemini app (Google)
- Content Credentials overview (Adobe Firefly Help)
- Frequently asked questions about Microsoft Designer (Microsoft Support)
- Microsoft announces signalling of generative AI content using IPTC and C2PA metadata (IPTC, 2023)
- Samsung Galaxy S25 series arrives worldwide (Samsung Newsroom)
- Working with manifests, remote manifests (Content Authenticity Initiative documentation)
- Verify Google AI-generated images and videos with SynthID (Gemini Apps Help)
- Pixel 10 and Content Credentials (Google)
- LinkedIn rolls out the C2PA standard (LinkedIn)
- Content Credentials on LinkedIn (LinkedIn Help)
- How this content was made: Captured with a camera (YouTube Help)
- C2PA welcomes TikTok to its Steering Committee (C2PA)
- Security Considerations 2.2 (C2PA)
- Verifying Provenance of Digital Media: Security Analysis of C2PA and its Implementation (IACR ePrint 2026/804)
- C2PA Cameras Do Not Survive Contact With Reality (David Buchanan, August 2026)
- Introducing the official Content Credentials icon (C2PA)
- C2PA User Experience Guidance 1.4 (C2PA)
- View Content Credentials (Adobe Help)
- Chrome browser extension overview (Adobe Help)
- Use the Chrome extension (Adobe Help)
- Find and recover matching Content Credentials (Adobe Help)
- Types of information in Content Credentials (Adobe Help)
- How we're increasing transparency for gen AI content with the C2PA (Google)
- Learn more about an image (Google Search Help)
- Apple Reference Image (Apple Security Research)
- Use Apple Reference Image (Apple Support)
- Sony's Authenticity Solutions (Sony)
- New: Leica M11-P (Leica Camera)
Free tools for this
Related guides
AI images
What is SynthID, and how do you check for it?
SynthID is the invisible AI watermark in Google and, since May 2026, OpenAI images. How it works, how to check it for free, and what a result proves.
AI images
How to tell if an image is AI-generated
Five steps that work: get the original file, read its labels, check for SynthID, look for visual clues and trace the source. What each test can prove.
Decision guide
Can you uncrop a photo?
Yes, if the app kept the original. Undo steps for iPhone, Mac, Google Photos, Samsung, Windows, Word and Canva, plus what to try when Revert is missing.