Content Credentials (C2PA): how to check where an image, video or audio clip came from.
Guide G12 · Published
For anyone checking an image, video or audio clip, including journalists. What Content Credentials are, how to read them, and why they are one input only: no tool, C2PA included, can declare content fake or real on its own.
Who this is for and when to use it
It is written for ordinary people, journalists, fact-checkers and communications teams who have received an image, video or audio clip and want to see whether the file carries information about its origin. It helps most when you have the file itself rather than a screenshot or a forwarded copy, and when it matters who signed the material and which edits are recorded.
It covers one step of checking, not the whole method. For the full sequence, see the guide “How to check a suspicious video, voice message or call”.
What Content Credentials and a manifest are
C2PA is an open technical standard that lets publishers, creators and consumers establish the origin and edits of digital content 1. Content Credentials is the name for the public-facing version: the C2PA explainer calls it the “preferred non-technical term” for a C2PA manifest 2. A manifest is a provenance record for a file that combines assertions, a claim and a signature 3.
- Assertions: individual statements about the file, such as its creation, editing actions and details of the capture device 3, and whether AI was involved 2.
- Claim: gathers the assertions together; the claim generator signs it on behalf of the signer 3.
- Digital signature: the device or software signs the manifest with its private key, and a verifier uses the matching public key 2.
- Binding to the content: hashes tie the manifest to the specific file 3.
The record is tamper-evident: any change to the file or the manifest breaks the hashes 2. That means a change can be detected, not that it is prevented: an inspection tool will show that the content was altered after signing 4.
What they can show
- Origin: who produced the material, when and where, with which device or software. This depends on what the creator or tool maker chose to record 4.
- Edits: which actions were applied and which ingredients the file was assembled from 4; whether AI was involved 2.
- The signer: the signer is identified by the subject of its credential 3. A creator need not include an identity, it may not be displayed, and the name you see may differ from the content’s owner 4.
- Changes after signing: if the content or its metadata was altered after it was signed, the inspection tool will show it 5.
Credentials answer “who signed this and what is recorded”, not “is this true”.
How to check
- 01Get the file closest to the original, from its author or the first source. Metadata is often lost when files are forwarded or uploaded to platforms 6.
- 02Open it in an inspection tool. A public option is the Content Authenticity Initiative’s Verify website (contentcredentials.org/verify) 4 5. If the material is confidential, read the tool’s privacy terms first.
- 03Read what it shows: who signed, which actions are recorded, whether an AI-involvement label is present, and whether the tool reports changes after signing 2 5.
- 04Judge the signer. A signature can be technically valid, yet anyone can create their own certificate; if the signer is not on a trust list, the software should make clear that the signer may not be who they claim to be 6. Ask yourself whether you know this signer and have reason to trust them.
- 05Record the result: the date, the tool, the file name and exactly what was shown. Then go on to the remaining checks: source, context and independent confirmation.
How to word the result
Keep to careful wording. For example:
- Credentials present, signer known and trusted: “Provenance is documented by signer X. This does not prove the content is accurate.”
- Credentials present, signer unknown or not on a trust list: “The signature is valid; the signer’s identity is not confirmed” 6 2.
- The tool reports changes after signing: “The file does not match its stored record; further checks are needed” 4.
- No credentials: “No Content Credentials found.” This proves nothing, because missing credentials are not, by themselves, a reason to distrust a file 2.
Limits of the method
- Not every file has credentials. Only tools and devices that support them add them, and edits made in tools without support may go unrecorded 2. Wide use also depends on tools supporting the standard and on users and platforms not stripping the metadata 6.
- They can be stripped. Metadata can be removed from a file 2; it is often removed when files are shared, for benign reasons such as privacy or to deceive recipients 6.
- Some approaches try to recover what was stripped. A soft binding (an invisible watermark or a fingerprint) can help find a manifest that is kept in a store 2 3. Joint guidance from cyber agencies in several countries treats such “durable” credentials as a way to protect media provenance 7. This works only where the watermark and the store actually exist.
- A missing credential proves nothing 2.
- A valid credential proves who signed, not that the content is true. The specification does not judge whether the recorded information is good or bad 3. NIST notes that transparency about origin does not guarantee trust and can create a false sense of it, for example when genuine content is taken out of context 6.
No tool, C2PA included, can declare content fake or real on its own. By its own authors’ account, Content Credentials “aren’t intended to prescriptively indicate whether a piece of content is ‘real’” 4. Detectors are no verdict either: NIST stresses that false positives, where human-made content is judged AI-generated, can cause serious reputational harm 6.
If you add credentials to your own material, find out what your tool records. Metadata can reveal when, where, by whom and with what the file was made 6.
Limits
This is a guide, not legal advice and not a service. NAK does not check files, does not sign or certify content, does not run a trust list and does not offer detection. The tools described belong to other organisations and change over time. This text was last checked on 24 Oct 2026; the sources used are listed below.
Sources
- 1C2PA. Coalition for Content Provenance and Authenticity (home page) · accessed 2026-10-24
- 2Coalition for Content Provenance and Authenticity. C2PA Explainer 2.2 · accessed 2026-10-24
- 3Coalition for Content Provenance and Authenticity. C2PA technical specification 2.2 · accessed 2026-10-24
- 4Content Authenticity Initiative. How it works · accessed 2026-10-24
- 5Content Authenticity Initiative. Frequently asked questions · accessed 2026-10-24
- 6NIST. Reducing Risks Posed by Synthetic Content: an overview of technical approaches to digital content transparency (NIST AI 100-4) · accessed 2026-10-24
- 7Canadian Centre for Cyber Security, with NSA, Australian Signals Directorate's ACSC and UK NCSC. Joint guidance: Content Credentials and strengthening multimedia trust in the generative AI era · accessed 2026-10-24