The C2PA standard: overview

The Coalition for Content Provenance and Authenticity, known as C2PA, is the dominant technical specification for embedding cryptographic provenance in digital media. As of mid-2026, C2PA is the only provenance specification with meaningful cross-industry adoption — present in cameras, phones, AI generators, editing software, and a small but growing set of consumer-facing browsers and platforms. This page explains what the coalition is, what the specification does at a high level, how it relates to adjacent efforts, and how it has evolved across its 2.x release line.

Three names recur and are routinely conflated. C2PA refers to the coalition and the technical specification it publishes. The Content Authenticity Initiative, or CAI, is a broader Adobe-founded community that promotes provenance adoption; CAI predates C2PA and is less formal. Content Credentials is Adobe's brand for the consumer-facing C2PA experience — the small CR icon, the inspection panel, the "Inspect Credentials" affordance. The standard is C2PA; the brand is Content Credentials; the community is CAI. The rest of this page uses these terms precisely.

The coalition

C2PA was formed in February 2021 as a project of the Joint Development Foundation, under the Linux Foundation. Founding members were Adobe, Microsoft, BBC, Arm, Intel, and Truepic. The coalition has expanded substantially since: by 2026 it includes the major US camera makers (Canon, Nikon, Sony, Leica), the major phone vendors (Google, Samsung, Apple as observer), the major image-generation platforms (OpenAI, Google, Anthropic on observer footing), large publishers (the New York Times, AP, Reuters), and a long tail of editing-software vendors. The specification itself is published openly at c2pa.org and developed in working groups under the JDF process.

The coalition does not operate as a certificate authority or trust-list maintainer by default; those functions sit alongside the specification work and are governed separately. The CAI handles much of the developer-relations and consumer-education effort. The split is functional: the coalition writes the spec, the CAI promotes its use, and a separate set of organizations operate the trust infrastructure.

What the specification defines

The C2PA Technical Specification, currently in its 2.x line, defines a family of related artifacts: assertions, claims, manifests, and the binding mechanisms that link manifests to media files. The architecture is summarized briefly here and covered in detail on the manifest structure page.

An assertion is a structured statement about an asset — a list of edit actions, an AI-generation flag, a capture device record, an ingredient pointer to a prior asset. Assertions are CBOR-encoded (RFC 8949) and stored in a manifest. A claim aggregates hashes of all assertions and the asset binding into a single signed object. The signature on the claim is what makes the whole structure tamper-evident. Manifests are stored inside JUMBF boxes (ISO/IEC 19566-5) embedded in the asset file format (JPEG, HEIF, MP4, PDF, WAV, and others).

The specification also defines the binding mechanisms — how a manifest is linked to the asset it describes. Hard bindings are cryptographic hashes of the asset bytes. Soft bindings are perceptual fingerprints or watermarks that survive moderate transformations. The combination is the basis for the durable Content Credentials mechanism added in the 2.x series.

What the specification does not define is what counts as a trusted signer. That decision is delegated to the C2PA Trust List, a separately governed registry of certificate authorities whose certificates are recognized by default. The trust list, the legacy Initial Trust List frozen on 1 January 2026, and the migration to the new governance process are covered on the trust list page.

Specification timeline

The 1.x to 2.x transition was significant. Several assertion types were restructured, the trust model was formalized, and the durable credential pattern was introduced. Validators built against 1.x continue to read 2.x manifests for most cases — the spec is conservatively backward-compatible — but producers targeting 2026 deployments are expected to use 2.x or later.

How C2PA relates to adjacent standards

C2PA does not exist in isolation. The JPEG Trust framework (ISO 22144), published in 2024 and refined in 2025, provides a reporting vocabulary for trust evaluations over C2PA manifests. JPEG Trust does not replace C2PA; it consumes C2PA and produces structured trust assessments. The integration is loose — a validator can implement both, or either — but in practice the two are converging in editorial and enterprise deployments.

The IPTC Photo Metadata Standard, maintained by the International Press Telecommunications Council, predates C2PA by decades and remains the dominant editorial metadata format. C2PA assertions can include IPTC fields verbatim, and the IPTC community published in 2023 a profile describing how its standards interoperate with C2PA manifests. For practical purposes, editorial photographers continue to use IPTC for descriptive metadata; C2PA wraps it cryptographically.

Adjacent watermarking standards — SynthID, Stable Signature, the IFRC's "Coalition for Content Identification" proposals — are complementary rather than competing. C2PA defines slots for watermark-derived signals to ride inside the manifest grammar, allowing watermarks to be cataloged and verified within the same provenance framework. The integration patterns are still being worked out, but the principle of layered defense is now consensus across the major actors.

Open governance and the trust question

The C2PA specification is open. Anyone can implement it; the c2patool reference implementation is open-source under the Apache license, hosted under the CAI's GitHub organization. The trust list is not as open: which CAs are listed, and the criteria for inclusion, are decisions made by C2PA member organizations under the JDF governance process. This split — open spec, gated trust — has been a source of recurring tension, especially among free-software and independent-publisher communities that want to issue their own signing certificates without relying on a coalition-blessed CA.

The C2PA position, articulated repeatedly in coalition documents through 2025, is that an untrusted certificate is still cryptographically validatable — the signature math works — and that the trust-list mechanism only governs default UI labeling. A self-signed C2PA manifest will validate; it will simply be marked "unknown signer" rather than "trusted." Whether this is sufficient depends on whether consuming applications give users meaningful tools to inspect untrusted signers, which is a UX question with no settled answer.

Where the field is moving

Through 2026, the dominant pressure on the C2PA specification is video. The 1.x line was built around still images; video was a later addition with several known integrity gaps around frame-level binding and live-stream signing. The 2.4 release closed some of these, but live-broadcast workflows in particular remain a research frontier. The major broadcasters in Project Origin have been pushing for video-first refinements that may produce a 3.0 release in the late-2026 to 2027 window.

The other live question is the consumer surface. Adobe, Microsoft, and Google have all shipped Content Credentials display surfaces, but the visual treatment varies. Whether browser vendors integrate a native C2PA badge — as Chrome was rumored to be evaluating in late 2025 — will determine whether ordinary readers ever encounter the credential as a routine signal or whether it remains a tool inspected only by editors, lawyers, and forensic analysts. The specification is mature; the social uptake is just beginning.