Short answer
C2PA is an open technical standard from the Coalition for Content Provenance and Authenticity for attaching signed provenance information to media files. Content credentials record how an asset was made and edited, including AI involvement. A DAM supports C2PA when it reads credentials on ingest, keeps them through every rendition and signs its own edits. Version 2.4 of the specification dates from April 2026.
What C2PA and content credentials are
C2PA is an open technical standard from the Coalition for Content Provenance and Authenticity for attaching signed provenance information to media files.
Content credentials are the signed provenance records attached to a media file under the C2PA standard. The specification uses "Content Credential" as the plain-language name for a C2PA manifest. Fact: the current release is the C2PA Technical Specification 2.4, dated April 2026.
This page is orientation for DAM buyers. It gives no legal advice on provenance or labelling duties.
| Building block | What it holds | Why a DAM team cares |
|---|---|---|
| Assertion | A statement about the asset, such as the actions taken or the digital source type | Carries the AI and edit history the DAM should display |
| Claim | The signed structure that lists the assertions and the content binding | Ties the statements to one exact file |
| Claim signature | The digital signature over the claim, made with the signer's private key | Proves who made the statements |
| Manifest | Assertions, one claim and one claim signature together | The unit the DAM reads, stores and passes on |
| Ingredient | An earlier asset used to make the current one | Links a rendition or composite to its master |
Why provenance matters for brand assets
Provenance records let a brand prove which assets are original, which were edited and which involved generative AI, with a signature that a third party can check.
Three uses matter for DAM buyers. Each one is a DAMSelection assessment of where provenance pays back in practice.
- Regulatory marking: the EU AI Act asks providers of generative systems to mark outputs in a machine-readable way. The Commission guidelines list cryptographic provenance methods among the accepted techniques, next to watermarks and metadata. Read EU AI Act obligations for AI-generated assets.
- Rights evidence: a signed record of creator and edits supports usage rights claims when an image is reused or disputed.
- Internal trust: teams downstream can check that a file in a feed is the approved version from the DAM.
Fact: the Commission guidelines on Article 50 do not name C2PA. C2PA is one available method, and the law sets no single mandatory standard.
How credentials attach to files
Content credentials attach to a file either embedded inside it or stored as an external manifest that the file references.
Fact: the specification lists JPEG, PNG, GIF and BMFF-based formats such as MP4 as embeddable. Version 2.4 adds HTML and structured text. Section 11 of the specification covers external manifests, which suit formats or workflows where embedding fails.
| Binding type | Mechanism | Strength | Weakness |
|---|---|---|---|
| Hard binding | Cryptographic hash of the asset or a part of it | Detects any change to the bytes | Breaks when the file is re-encoded without a new manifest |
| Soft binding | Fingerprint or invisible watermark | Survives metadata stripping, so the manifest can be found again | Not unique, so the specification forbids its use as a hard binding |
Fact: a manifest may contain only one hard binding assertion. DAMSelection assessment: DAM teams should expect both types in practice, because CDNs and social platforms often strip embedded metadata.
Recording AI involvement
C2PA records AI involvement through the digital source type, which uses the IPTC Digital Source Type vocabulary.
The C2PA guidance for AI and machine learning asks generators to identify AI output with the right digital source type. Version 2.4 also adds an AI disclosure assertion. The values below come from the IPTC vocabulary and map to the AI fields in your metadata schema.
| IPTC value | IPTC label | Typical DAM example |
|---|---|---|
| digitalCapture | Digital capture sampled from real life | Studio product photo |
| computationalCapture | Multi-frame computational capture | Smartphone HDR image |
| algorithmicallyEnhanced | Algorithmically-altered media | Sharpened or noise-reduced photo |
| compositeWithTrainedAlgorithmicMedia | Edited using generative AI | Photo with an AI-extended background |
| trainedAlgorithmicMedia | Created using generative AI | Fully generated lifestyle scene |
| compositeSynthetic | Composite including generative AI elements | Campaign key visual with generated elements |
DAMSelection assessment: store the digital source type as a controlled field in the DAM as well as in the manifest. Search, filters and publishing rules then work even on files whose manifest got stripped.
What DAM support looks like
DAM support for content credentials ranges across four levels, and buyers should ask which level each vendor reaches for each file format.
The levels are a DAMSelection assessment for scoring. Test them on your own files in the demo, because support often differs between images and video.
| Level | What the DAM does | What breaks without it |
|---|---|---|
| 1 Display | Reads the manifest on ingest and shows it in the asset detail view | Users cannot see provenance at all |
| 2 Preserve | Keeps the manifest on the master and on generated renditions | Credentials vanish at the first resize |
| 3 Sign | Adds its own signed manifest for edits made in the DAM, with the earlier manifest as ingredient | The edit history stops at upload |
| 4 Validate | Checks signatures against the C2PA Trust List and flags invalid or unknown signers | Forged or expired credentials pass unnoticed |
Fact: the specification treats re-encoding and scaling as non-editorial actions that produce an asset rendition. Version 2.3 added a proportional resizing action and clarified the converted action. A level 3 DAM records those actions when it creates renditions through dynamic media or batch jobs.
Where DAM workflows break credentials
Content credentials break in a DAM wherever a process rewrites the file without writing a new manifest.
DAMSelection assessment from PIMvendors practice: these five points cause most losses. Map each one in the integration design and test it before go-live.
- Rendition generation that re-encodes images without carrying or re-signing the manifest.
- Image CDN delivery that strips metadata to save bytes.
- Exports to marketplaces and social platforms that accept only plain files.
- Creative round trips, where a designer opens the asset, edits it and saves a new version.
- Metadata write-back of XMP or IPTC fields, which changes the bytes and invalidates the hard binding.
The specification allows redaction of some assertions, with a record of the removal in the claim. Fact: claim generators may not redact the actions assertions. Ask vendors how they handle a request to remove personal data from a manifest.
Trust list and conformance
The C2PA Conformance Program checks generator and validator products against the specification and lists conforming products publicly.
Fact: the official C2PA Trust List launched in mid-2025 under the Conformance Program. The earlier Interim Trust List was frozen on 1 January 2026, and content signed under a then-valid interim certificate stays valid under the legacy model. DAMSelection assessment: ask whether the vendor's signing certificate sits on the C2PA Trust List, and whether the vendor appears on the conforming products list.
Where C2PA fits in the scoring model
C2PA support touches four of the twelve DAMSelection scoring criteria, so score it inside each of those four criteria.
The mapping is a DAMSelection assessment. Weight it by how much generated or externally sourced content flows through your library, and record the level each vendor proves in the demo.
| DAMSelection criterion | C2PA question to score |
|---|---|
| Asset ingestion and processing | Does ingest read and keep manifests for every format you use |
| Distribution and CDN | Do delivered renditions carry valid credentials or a soft binding |
| Creative tool integration | Do round trips from design tools keep the ingredient chain |
| Security, hosting and compliance | Does the DAM sign with a trusted certificate and validate incoming signers |
DAMSelection assessment: buyers with few generated assets can accept level 2 today and write level 3 into the roadmap clause of the contract. Buyers producing generated campaign imagery at scale should require level 3 at go-live.
Questions for vendors
Vendors should answer C2PA questions in writing and prove the answers on the buyer's files in a scripted demo.
Add these to your scripted DAM demo or generate them with the Demo Script Generator.
- Upload a camera file and a generated image with credentials, and show the manifests in the asset view.
- Generate a web rendition and a crop, then validate both files in a public C2PA verifier.
- Edit the asset inside the DAM and show the new signed manifest with the earlier one as ingredient.
- Show which file formats support embedded manifests and which use external manifests.
- Name the signing certificate authority and its trust list status.
- Show how credentials appear in search filters and in the API response.
For the bigger picture, see AI in DAM and the DAM Learn pillar. The glossary covers watermarking and renditions.
Frequently asked
What is C2PA in digital asset management?
C2PA in digital asset management is the use of the C2PA standard to read, keep and write signed provenance records on assets. A DAM with C2PA support shows each asset's creation and edit history, keeps credentials on renditions and signs edits made inside the platform. Support levels differ by vendor and file format.
Are content credentials the same as a watermark?
Content credentials are signed manifests that bind statements to a file through a cryptographic hash. A watermark is a mark hidden in or shown on the pixels. C2PA uses invisible watermarks and fingerprints as soft bindings, which help recover a manifest after metadata stripping. Many workflows combine both methods.
Do content credentials survive resizing?
Content credentials survive resizing only when the system that resizes the file writes a new manifest that references the original as ingredient. A plain resize changes the bytes and breaks the hard binding. Test every rendition path in the DAM and CDN with a C2PA verifier before go-live.
Which C2PA version should a DAM support?
A DAM should support the current C2PA Technical Specification, version 2.4 from April 2026, and read manifests written under earlier 2.x versions. Ask vendors for their roadmap to new releases, because the specification adds actions and assertions in each version, such as the AI disclosure assertion in 2.4.
Does the EU AI Act require C2PA?
The EU AI Act requires machine-readable marking of generative AI outputs under Article 50(2), but names no single standard. The Commission guidelines list watermarks, metadata, cryptographic provenance, logging and fingerprints as techniques. C2PA content credentials are one way to meet the duty.
Sources
- C2PA specifications
- C2PA: Coalition for Content Provenance and Authenticity
- C2PA Specifications 2.4 index
- C2PA Content Credentials Technical Specification 2.4
- C2PA Guidance for Artificial Intelligence and Machine Learning 2.3
- C2PA Conformance Program and Trust List
- IPTC NewsCodes: Digital Source Type
- Commission guidelines C(2026) 5054 on Article 50 transparency obligations
- AI Act Service Desk: Article 50, consolidated text