ProvenanceGuard
← Go back

← All guides

For AI Image Providers: Testing Whether Your Outputs Stay Marked

By the Ingenious AI team · Published: November 14, 2026 · Reviewed: November 14, 2026 · 8 min read

The short answer

Register each system version in the Evidence Vault as it ships, run the battery on genuine sample outputs, and keep the hash chain unbroken across versions. Free covers a first sample battery per system; the paid depth is the continuing history and exportable packs. Your customers' Article 50 questions then meet a chain head and measured runs, not a blog post about your commitment to transparency.

Seal an image free → Recover a reposted image Pricing

Providers sell marking as a feature in a market where buyers are learning to ask for measurements; the vault is how a small provider answers like a big one.

The Evidence Vault answers a provider-side question: does this generative system's output stay machine-readably marked after a realistic pipeline touches it? A system is registered in the vault, a sample output is uploaded, and the conformance battery runs real transforms on the real bytes: re-encoding, resizing to half size, format conversion, and metadata stripping. After each transform the bytes are re-inspected for marking, C2PA and JUMBF markers, XMP, a ProvenanceGuard assertion, or AI disclosure in metadata, and the before-and-after results are recorded as measured, never assumed.

What is provider marking practice? the routine by which a generative system's provider registers each version, runs marking-conformance batteries on its real outputs, and maintains a signed history that customers and reviewers can verify.

Why provenance for AI providers matters

Each battery run is sealed into a hash chain: the run's pack hash covers the previous run's pack hash plus the canonical record of this run, and the pack is HMAC-signed. Tampering with any stored field breaks recomputation, and the chain can be checked through the vault's API. That structure is what turns a folder of test results into evidence: a provider preparing for the Article 50(2) marking duties can show not just a passing run, but an unbroken, signed history of runs across versions. The free tier includes one sample battery per system; conformance history across repeated runs is a Pro and Team feature, as is evidence-pack export.

The legal backdrop in the product's own evidence wording is specific. The EU AI Act's Article 50 transparency duties have applied since 2 August 2026, and the marking grace period for systems already on the market before that date ends 2 December 2026. California's AI Transparency Act, SB 942, is operative from 2 August 2026. New York's synthetic performer law, in GBL section 396-b and effective 9 June 2026, covers disclosure where commercial advertising features a synthetic performer. The evidence packs carry disclosure wording for exactly these three jurisdictions.

How the record is built

None of that makes a provenance record a legal verdict, and ProvenanceGuard's packs say so themselves: evidence aid, not legal advice. What a record does is give counsel, a compliance team or a regulator something concrete to work from: the file, its hashes, the tool that sealed it, the timestamps, the declared disclosure and the wording used. Organisations that can produce that pack in minutes answer questions very differently from organisations reconstructing a publication history from chat logs and memory.

The practical reading of these dates is about preparation rather than panic. Duties that have already applied cannot be met retroactively for assets published unmarked, but a workflow adopted now, applied consistently to everything published from here on, with records kept per asset, is a defensible position that improves with every week it runs. The grace deadline in December 2026 is specifically about marking for pre-existing systems, which makes conformance testing of the marking itself, not just the policy around it, the urgent piece of work.

Where it meets the real world

An evidence pack is the record, packaged for a reader who was not there. For a sealed asset it contains the exhibit ID and verification URL, the original and sealed SHA-256 hashes, the perceptual fingerprint, the file's format and dimensions, the tool identification and version, the registration timestamp, the creator and disclosure as declared, the inspection results for the original bytes, and disclosure wording by jurisdiction for the EU, California and New York. Exporting the pack is a Pro and Team feature, because it is the point where a working record becomes a compliance artefact.

The pack's value is mostly about timing and completeness under pressure. A brand safety question, a platform appeal, a client audit or a regulator's letter all arrive with deadlines, and all of them ask the same underlying questions: which file, made how, declared as what, published when. A team that exports the pack and sends it the same day looks, and is, organised. A team assembling screenshots and recollections does not, however innocent its actual process was. Evidence does not create innocence, but it is how innocence is demonstrated.

Putting it into practice

Provenance is the record of where a piece of content came from and what has happened to it since: who or what created it, when it was registered, and which copies descend from the original. For an AI-generated image that record matters because the file itself is easy to copy, rename, crop and re-upload, and none of those actions leaves a reliable trace inside a filename. A provenance record kept outside the file, in a registry keyed to the file's fingerprints, survives exactly the handling that erases the signals embedded in the file.

The practical problem is distribution. An image is rarely published once, in one place, in its original form. It is uploaded to platforms that resize and re-encode it, forwarded by partners, saved and reposted by audiences, and archived by people who strip whatever context surrounded it. Each step is normal and usually innocent, yet together they separate the image from anything that said how it was made. Planning for that separation, rather than hoping it will not happen, is the whole discipline this guide is about.

Five actions to take

  1. Create the vault system entry at release time, with the version label exact
  2. Run the battery on outputs your customers would actually receive
  3. Seal the final export of the asset, with the creator and disclosure declared in your own accurate words
  4. Download and keep both copies the seal produces: the sealed copy as the record, the labelled copy for publication
  5. File the exhibit's verification URL beside the asset wherever your team keeps campaign or client records

A worked example: A two-person image generation startup in diligence with an enterprise buyer

The buyer's technical review recomputed the startup's vault chain, 14 runs over 5 versions, all valid, and marking survival results were pasted from the pack into the diligence memo, closing that workstream in a day.

Notice what made the difference: not a better argument, but a record created before it was needed, carrying dates and hashes a stranger could check. That is the whole pattern this guide recommends, applied at the scale of one team and one deadline.

FAQ

What should a provider publish about its marking?

The measured results and the chain's verifiability, with the method's limits stated: which transforms, which detections, which versions. Published precision invites verification; published adjectives invite the other kind.

Is ProvenanceGuard's assertion a C2PA manifest?

No. It is ProvenanceGuard's own signed assertion, embedded as XMP and HMAC-signed, and labelled that way everywhere. The inspection detects genuine C2PA markers when a file carries them, but the two records are never conflated.

What formats and sizes are supported?

JPEG and PNG, up to 25 MB per file. Those are the formats the inspection, XMP embedding and fingerprinting are built and tested against; other formats are refused rather than approximated.

What does the free tier actually do?

The complete job on an asset: byte-level inspection, a signed assertion embedded as XMP, a visibly labelled copy, SHA-256 and perceptual fingerprints registered, and a public verification page. Paid plans add survival re-checks, evidence packs and bulk verification.

Seal an image free →