What is C2PA?
C2PA is two things that share a name: a coalition of companies, and the technical standard that coalition writes. The standard defines how software and hardware can attach a signed record to a media file stating where the file came from and what has been done to it. C2PA describes itself as a project of the Joint Development Foundation, a nonprofit standards body, focused on open, global technical standards for content provenance (C2PA).
The record is tamper-evident, not a judgment about truth. C2PA's own explainer says Content Credentials do not provide value judgments about whether a given set of provenance data is true, and calls the standard no cure-all for misinformation (C2PA explainer). What a valid credential shows is that a particular signer attached particular statements to a particular file, and that neither has changed since.
C2PA matters for compliance because laws in the European Union and California now require machine-readable provenance signals on AI-generated content. Our analysis of watermarking and C2PA covers that argument; this page explains the standard itself.
C2PA meaning and who is behind it
C2PA stands for Coalition for Content Provenance and Authenticity. It was announced on February 22, 2021 by six founding members, Adobe, Arm, BBC, Intel, Microsoft and Truepic, and merged two earlier efforts: the Adobe-led Content Authenticity Initiative and Project Origin, led by Microsoft and the BBC (Microsoft announcement).
As of November 12, 2026, the C2PA membership page lists 11 steering committee members: Adobe, Amazon, BBC, Google, Meta, Microsoft, OpenAI, Publicis Groupe, Sony, TikTok and Truepic. Arm, a founder, now appears among the general members.
Three names are easy to confuse:
- C2PA is the standards body and the specification.
- Content Credentials is the public-facing name. The specification calls Content Credential the preferred non-technical term for a C2PA Manifest (C2PA specification 2.4).
- The Content Authenticity Initiative is a separate community that promotes adoption and provides open-source tools and software kits for implementing the standard (Content Authenticity Initiative).
How the C2PA standard works
A C2PA-enabled camera, editing program or AI generator writes statements about a file, bundles them, signs the bundle with a certificate, and stores the result in or alongside the file. A validator later checks the signature, checks that the file still matches, and checks whether the signer is on a trusted list. The table summarizes the parts, using definitions from the C2PA specification 2.4.
| Component | What it does | Defined in | Notes |
|---|---|---|---|
| Assertion | A statement about the asset, made by the signer or gathered when the claim is generated | Specification glossary | Examples: actions taken, ingredients used, AI disclosure |
| Claim | A signed, tamper-evident structure that references the assertions and the content binding | Specification glossary | One claim per manifest |
| Claim signature | The digital signature on the claim, made with the signer's private key | Specification glossary | Only X.509 certificates may sign, since version 2.0 |
| C2PA Manifest | The assertions, the claim and the claim signature together | Specification glossary | Also called a Content Credential |
| Manifest Store | The collection of manifests for an asset | Specification glossary | Can be embedded in the file or kept outside it |
| Hard binding | Cryptographic hashes that uniquely identify the asset or part of it | Specification, content bindings | How edits to the file are detected |
| Soft binding | A fingerprint or invisible watermark that identifies the content | Specification, content bindings | Used to find a credential again after metadata is removed; may not replace a hard binding |
| C2PA Trust List | A C2PA-managed list of certificate trust anchors for hardware and software signers | Specification glossary | Published in C2PA's public conformance repository |
C2PA manifest
The manifest is the unit of provenance. Each time a conforming tool creates or edits a file it can add a new manifest, and the set of manifests travels as a Manifest Store. The last one in the store, whose content bindings can be validated against the file as it now exists, is the active manifest. When a new file is built from existing ones, the earlier files are recorded as ingredients through an ingredient assertion, which is how an edit history is chained together (specification 2.4).
For AI content, the relevant statement is the action's digital source type. C2PA's guidance says that for a generative model the value will be trainedAlgorithmicMedia (C2PA guidance for AI and machine learning). Version 2.4 added a dedicated AI Disclosure Assertion, labelled c2pa.ai-disclosure, for machine-readable AI transparency information.
C2PA metadata and signing
C2PA data is metadata: structured information stored with the file, not pixels or sound. The claim is encoded in a JUMBF box, a metadata container format from the JPEG family, and signed. C2PA's FAQ describes the cryptographic base as SHA-256 hashes, X.509 certificates and digital signatures (C2PA FAQ). Trusted time-stamps let a validator confirm that a signature was made while the signing certificate was valid, even if the certificate has since expired.
Because it is metadata, it can be removed. The C2PA explainer answers the question directly: can the provenance metadata be removed? Yes, it can (C2PA explainer). The standard's response is the durable Content Credential: a credential with one or more soft bindings, such as an invisible watermark or a fingerprint, that allow it to be found again in a manifest repository after the embedded copy is gone.
C2PA trust list
A valid signature only proves that someone signed. Whether that signer should be trusted depends on the C2PA Trust List, which the C2PA FAQ describes as a curated list of certification authorities authorized to issue signing certificates for products that generate credentials (C2PA FAQ). A second list, the TSA Trust List, covers time-stamping authorities. Both are published in C2PA's public conformance repository.
The trust list is tied to the C2PA Conformance Program, which opened in June 2025 and is now at version 0.2. C2PA says the program holds generator products, validator products and certification authorities accountable to the specification, its certificate policy and its security requirements (C2PA conformance). The Content Authenticity Initiative reported on October 9, 2025 that Google's Pixel 10 was the first device to reach the program's highest conformance level (Content Authenticity Initiative).
An earlier Interim Trust List was frozen on January 1, 2026: no new certificates are added, and existing ones remain valid for legacy support (Content Authenticity Initiative trust list documentation).
C2PA specification versions
Dates and descriptions below are from the version history inside the specification itself, except version 2.5, which is listed in the 2.5 specification's contents.
| Version | Date | Main changes |
|---|---|---|
| 1.0 | December 2021 | Initial release |
| 1.1 | September 2022 | Not summarized here |
| 1.2 | October 2022 | Added the digital source type field to actions |
| 1.3 | April 2023 | Second versions of actions and ingredients; WebP, AIFF, AVI and GIF |
| 1.4 | November 2023 | ZIP-based formats and fonts |
| 2.0 | January 2024 | Trust shifted from identifying people and organizations to hardware and software identified by X.509 certificates; C2PA Trust List introduced |
| 2.1 | September 2024 | Security fixes; TSA Trust List; improved soft bindings; JPEG XL; edits to prepare for ISO standardization |
| 2.2 | May 2025 | Soft Binding Resolution API; multi-part assets; trust list limited to claim-signing certificates |
| 2.3 | December 2025 | Live video streaming; manifests in unstructured text files; finer watermarking actions |
| 2.4 | April 2026 | AI Disclosure Assertion; manifests in HTML and structured text |
| 2.5 | October 2026 | Current version; we have not found an official summary of its changes |
C2PA's FAQ says implementations of the 1.x series will be considered legacy because the 2.x series made significant security improvements (C2PA FAQ). The standard is also on a path to ISO. ISO's catalogue lists a draft titled "Authenticity of information: Content credentials" under the number 22144 as under development, with no publication date (ISO catalogue record, checked November 12, 2026). It is not accurate to call C2PA an ISO standard yet.
Content provenance and the law
Content provenance means information about the origin and history of a piece of content. Two legal regimes now require it for AI output, and neither names C2PA in the text we reviewed.
EU AI Act Article 50
Article 50(2) of Regulation (EU) 2024/1689 requires providers of AI systems that generate synthetic audio, image, video or text to ensure outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. Their technical solutions must be effective, interoperable, robust and reliable as far as technically feasible (European Commission AI Act Service Desk). The transparency obligations apply from August 2, 2026, and the Commission published a final, voluntary Code of Practice on Transparency of AI-generated Content on June 10, 2026 (European Commission).
The Digital Omnibus on AI gave generative systems placed on the market before August 2, 2026 until December 2, 2026 to meet the marking duty, as described in Gibson Dunn's alert on the agreement. Details are on our EU AI Act page and the older AI Act content rules page.
California AI Transparency Act
California's act requires covered generative AI providers to embed latent disclosures in the content their systems produce, operative August 2, 2026. From January 1, 2027, large online platforms must detect provenance data that is compliant with widely adopted specifications adopted by an established standards-setting body, and may not, to the extent technically feasible, knowingly strip such data or digital signatures (Cal. Bus. & Prof. Code § 22757.3.1). That phrasing describes standards like C2PA without naming one. See our California AI Transparency Act page for the full schedule.
What C2PA does and does not cover
C2PA gives a provider a standardized, signed, machine-readable way to say that a system generated a file. It does not, by its own account, survive every removal: metadata can be stripped, and recovery depends on soft bindings. Whether a given implementation is robust and reliable enough for Article 50(2), or carries every field California requires, is a question about that implementation and the legal text, not something the standard certifies. Other marking methods are covered on our compliance hub.
How to verify C2PA data
The official checking tool is the C2PA Verify site at verify.contentauthenticity.org. A user uploads a file and the tool reads any Content Credentials in it. If the credential was signed with a known certificate, the tool displays the name of the certificate owner and the time of the claim signature; if not, it reports an unknown source (Content Authenticity Initiative documentation).
Three cautions apply when reading a result:
- No credential proves nothing. The C2PA explainer says no assumption should be made about the trustworthiness of an asset purely based on whether it uses Content Credentials (C2PA explainer). Most files have none, and credentials can be removed.
- History can be incomplete. The explainer states that provenance is not always complete.
- The trust list in use matters. The same documentation says the Verify tool currently validates against the frozen Interim Trust List and will be updated to use the C2PA trust lists.
Independent researchers have also criticized the design. A 2026 preprint by Enis Golaszewski, Neal Krawetz, Alan T. Sherman and co-authors, which analyzed specification 2.2 and version 0.1 of the conformance program, concluded that the specifications and conforming implementations failed to achieve claimed security goals, and said some of its suggested improvements were adopted in version 2.3 (IACR ePrint 2026/804). The paper is not peer-reviewed, and both versions it studied have been superseded.
C2PA is an open standard for signed provenance metadata, published by a coalition founded in 2021 and now at specification version 2.5, dated October 2026. A valid credential shows who signed a file's history and that it has not changed; it does not show that content is true, and it can be removed. As of November 12, 2026, the EU and California texts we reviewed require machine-readable provenance without naming C2PA.
What changed
- 2026-11-17: Page published, with specification and trust list status checked November 12, 2026.
- 2026-01-01: Interim Trust List frozen.
Frequently asked questions
What is C2PA?
C2PA is the Coalition for Content Provenance and Authenticity, a standards project founded in 2021 by Adobe, Arm, BBC, Intel, Microsoft and Truepic. It publishes an open specification for attaching signed, tamper-evident provenance records, called Content Credentials, to images, video, audio and documents.
What is a C2PA manifest?
A C2PA Manifest is the set of provenance information for an asset: one or more assertions about the file, a single claim that references them, and a digital signature on that claim. Content Credential is the non-technical name for the same thing. Manifests can be embedded in a file or stored outside it.
Is C2PA a watermark?
No. C2PA data is signed metadata stored with a file, and C2PA says it can be removed. The standard lets an invisible watermark or fingerprint act as a soft binding that helps recover a credential after the metadata is stripped, but the watermark is a companion to the credential, not the credential itself.
Does C2PA satisfy AI labeling laws?
Not automatically. EU AI Act Article 50(2) requires machine-readable marking that is effective, interoperable, robust and reliable, and California refers to widely adopted specifications from an established standards-setting body. Neither text we reviewed names C2PA, so compliance depends on how a specific implementation meets each law.
Sources
- C2PA: Content Credentials technical specification 2.4
- C2PA: technical specification 2.5
- C2PA explainer
- C2PA: about
- C2PA: FAQ
- C2PA: membership
- C2PA: conformance program
- Content Authenticity Initiative: trust list documentation
- Microsoft: C2PA founding announcement (Feb. 22, 2021)
- European Commission AI Act Service Desk: Article 50
- European Commission: Code of Practice on Transparency of AI-generated Content
- Cal. Bus. & Prof. Code § 22757.3.1 (Justia)
This page is journalism and commentary, not legal advice. AI law changes quickly; check the primary sources above and consult qualified counsel for your situation.
