Metadata and provenance

Content Credentials explained for photo archives

A soft toy and a passenger look out of a car window beside an open coastal road.
A travelling companion.Photograph by Robert Harper. Robert Harper Photograph Library, Archive.London. From the archive, shown for illustration.

Content Credentials can prove who signed a file and whether it has changed since. They cannot prove that a caption is true, and for a film archive that distinction decides where they are worth using.

A Content Credential answers two questions about a file: who signed it, and has anything changed since. It does not answer the question most people actually want settled, which is whether the picture shows what its caption says. Keeping those apart is the whole skill of using the technology well, and it matters especially for archives, where the event in the photograph and the creation of the digital file may be fifty years apart.

The status of the standard and its adoption below was checked on 17 November 2026.

What are Content Credentials, and what is C2PA?

The Coalition for Content Provenance and Authenticity (C2PA) is the standards body, a Linux Foundation project that marked its fifth anniversary in February 2026. Content Credentials is the public name for what its specification defines. The specification's version history dates version 2.4 to April 2026; the C2PA announced version 2.3 on 9 February 2026. A draft has also been through the ISO process as ISO/DIS 22144, though whether the International Standard has been published could not be confirmed for this article.

The unit of the system is the manifest. It has three parts:

  1. Assertions: statements about the asset, such as the capture device, the software used, the actions performed (created, opened, edited) and the digital source type. One of them is the hard binding, a cryptographic hash of the content that ties the record to these exact bytes.
  2. A claim: the assertions gathered into a single statement ready for signing.
  3. A claim signature: the claim signed with the signer's private key, using an X.509 certificate.

When a signed file is edited in software that supports the standard, the earlier manifest is kept as an ingredient of a new one, so the history accumulates. In a JPEG the manifests are stored in a JUMBF box inside the file. The specification also allows a soft binding, an invisible watermark or fingerprint that lets a credential held elsewhere be matched to a copy whose embedded data has gone.

Who signs, and why should anyone trust the signature?

Trust in the system rests on the signer's certificate. The specification is explicit that trust decisions attach to the identity bound to the signing key, not to identifying an individual human. A credential says "a device or service holding this certificate signed this", and a validator checks whether that certificate chains to an entry on the C2PA Trust List.

That list, and the conformance programme that tests products which generate and validate credentials, launched in mid-2025. The earlier interim list was frozen on 1 January 2026.

This gives three levels of result, and they should not be confused: a manifest can be well-formed (structurally correct), valid (the hash matches and the signature checks), and trusted (the signer is on a recognised list). A self-signed credential can be perfectly valid and still tell you only that somebody signed it.

How do you check whether an image has Content Credentials?

The quickest first look is the free Content Credentials checker, which runs in the browser, detects whether a JPEG or PNG contains a C2PA manifest and lists what can be read from it, including the claim generator and whether AI-generation assertions appear. It detects and displays. It does not validate signatures cryptographically, and for that you should use the official Verify tool at contentcredentials.org.

Two reading rules:

  • Absence means nothing. Most photographs in circulation have no credential, and files that once had one often lose it in transit.
  • Presence means what it says, no more. Read who signed and what they asserted before drawing any conclusion.

For the ordinary descriptive and rights fields, which are separate from C2PA, use the EXIF and metadata viewer.

How widely are Content Credentials used in late 2026?

More than headlines suggest in hardware, less than they suggest in publishing. The following is drawn from the C2PA's own announcements and from a secondary survey published in March 2026 and updated in September 2026.

Area Position as reported
Membership The C2PA said in July 2026 that it has more than 500 members; TikTok joined its Steering Committee on 28 July 2026
Phones Google's Pixel 10 signs every photograph by default; Samsung's Galaxy S25 was reported to attach credentials only to images created or edited with generative AI
Cameras Leica, Sony and Nikon have produced capable cameras; Nikon suspended its signing service and revoked certificates after a vulnerability, and it had not been restored in early 2026
Platforms LinkedIn displays a credential icon; Cloudflare Images can preserve credentials if the site owner opts in; many social upload pipelines still strip them
Search Google documents that it can extract C2PA metadata for its "About this image" feature
News publishing The Reuters Institute's 2026 trends report put the share of news images and videos published with C2PA metadata at under 1%

The Nikon episode is a useful reminder that the system depends on key security. When signing keys are compromised, certificates are revoked and credentials issued under them stop being trusted.

Much of the momentum in 2026 has come from the need to label synthetic media. On 31 July 2026 the C2PA published guidance on using credentials to identify synthetic and non-synthetic content, built around the IPTC Digital Source Type vocabulary and a new AI disclosure assertion.

What Content Credentials cannot do

  • They cannot verify the scene. A credential attests to the file's handling. A staged photograph signed in camera is still staged.
  • They cannot survive every pipeline. Embedded manifests are removed by many platforms and by any tool that re-encodes without supporting the standard. Soft bindings are the standard's answer, and they depend on someone operating a lookup service.
  • They are not a rights reservation. On 22 January 2026 the C2PA stated that its specification contains no standard assertion relating to text and data mining and that the subject is out of scope. The file-level signal for that purpose is described in signalling AI data-mining preferences in image metadata.
  • They do not establish ownership. A signature shows who vouched for a file. Copyright ownership is a matter of documents and law.
  • They are not required for authenticity. The whole of twentieth-century photography predates them.

What can a credential say about a scan of a negative?

This is where archives need to be exact. A photograph taken on film in the 1970s cannot acquire a capture-time credential. What can be signed is the digitisation. A manifest created when a transparency is scanned or when the master file is exported could record that the named signer produced this file on this date, with a digital source type of positiveFilm or negativeFilm, and every later derivative could carry that manifest forward as an ingredient.

That is a real gain: it fixes the start of the digital chain to an accountable party, and it lets a picture desk distinguish the archive's own scan from a copy of unknown origin. But it starts the clock at the scanner. Everything earlier rests on the physical original, the sleeve, the contact sheet and the paper record, which is the subject of provenance for photographs.

A simpler measure complements it. Recording a SHA-256 hash of each master scan at the moment of creation, which the image hash and fingerprint tool can compute locally, gives an archive a private register against which any later copy can be tested for exact identity. It needs no certificate and loses nothing when metadata is stripped, because it is kept outside the file.

Should an archive adopt Content Credentials now?

For most archives the honest answer is: plan for it, do not wait for it. Get the foundations right first, meaning stable identifiers, complete rights fields and a hash register. Ask scanning and asset-management vendors whether their software can sign on export and whether their signing certificates are on the C2PA Trust List. Sign delivery files once the tooling is in place, and expect many recipients to see no credential at all until platforms preserve them more reliably.

Where the Robert Harper archive stands

The Robert Harper Photograph Library consists of 95,487 unique photographs by its inventory count, not yet checked frame by frame against the physical material, and its 52,684 analogue images largely await digitisation. The collection catalogued as David Bowie photographs is the kind of material where a picture desk will want to know that a file came from the rights holder's own scan. Because scanning is still ahead, provenance can be designed into the workflow from the first frame. Agencies and publishers with delivery requirements, including Content Credentials, are invited to raise them through the page for picture agencies, and the full set of browser utilities is listed under free image tools.

Frequently asked questions

What are C2PA Content Credentials?

They are tamper-evident records attached to a media file under the open C2PA specification. A credential lists statements about the file, such as the device or software that produced it and the edits made, binds them to a hash of the content, and is signed with a certificate. A verifier can then tell whether the file or the record has been altered since signing.

How do I check whether an image has Content Credentials?

Open it in a tool that reads C2PA manifests. A browser-based checker can detect a manifest and show its contents without uploading the file. To validate the signature against the C2PA Trust List, use the official Verify tool at contentcredentials.org. A missing credential proves nothing: many genuine photographs have none, and many platforms remove them.

Do Content Credentials prove a photo is real?

No. They prove that a particular signer made particular statements about a file, and that the file has not changed since. A camera that signs at capture gives strong evidence about that capture. A credential added later by software or an archive shows who vouches for the file from that point on, not what the picture depicts.

Do Content Credentials stop AI training on my images?

No. Content Credentials record provenance and do not reserve rights. The C2PA stated on 22 January 2026 that its specification has no standard text and data mining assertion and that the subject is out of scope. A rights reservation needs a different signal, such as the IPTC Data Mining property or a site-level protocol.

Can scans of old negatives carry Content Credentials?

Yes, if the scanning or export software can sign. The credential would record the digitisation: who signed, when, and that the source was film. It would be the first link in the digital file's history. The decades before the scan still have to be evidenced by the physical original and the archive's paper records.

Sources

Published by Archive.London. Corrections are welcome under our editorial policy.

For picture agencies, publishers and researchers

Fashion, beauty, jewellery, music, portraits and British life from the 1960s onwards, with 52,684 analogue originals still to be scanned. Tell us what you are researching.

See representation and researchWrite to the archive