Content-Provenance & AI-Disclosure AI Agent

What this agent does

Assembles the file, not the label a person signs

An asset is generated or altered, and the tool, the vendor and the transform are read from the pipeline record.

An asset arriving with a manifest has each assertion validated as intact — tamper-evidence, not a truth claim.

Reason

A market is added, and the asset is tested against the marking and disclosure duties that market imposes.

An asset is matched to Art. 50(2) marking and to Art. 50(4) deployer disclosure — both can attach to one asset.

An asset claimed under the assistive-editing carve-out carries a written reason the semantics were untouched.

Decide

An asset is re-checked at packaging, and the transform where its provenance was lost is named.

An asset inside an evidently artistic work routes for a disclosure of existence, which is a reduced duty.

Out

An asset is held with its manifest, its evidence, the residual ambiguity and the label proposed for it.

Execute write actions only inside the approval boundaries agreed during implementation.

→ Product statement

The agent assembles and classifies; a named compliance lead publishes the label and answers for it, and the company stays the deployer.

Example workflow

One asset, ingest to label

AgentHuman

1Asset receivedPipeline record, vendor manifest, transform log or delivery package

2Provenance assembledTools, transforms and assertions, each with where it survived and where it did not

3Classification proposedMarking duty, disclosure form, market and confidence

4Controls appliedManifest-validity checks, per-market duty checks, carve-out reasoning and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing is labelled at any of them — the agent is assembling, and the lead's lane opens at the confidence gate.

5DecisionBranches at the confidence threshold

High confidence

Goes to the compliance lead to approve.

Low confidence

Adds a legal read first.

Compliance approval

The file is held with its manifest, its residual ambiguity and the confidence.

Approve · Amend · Send to legal review

Approved — released to label▼

6Publishing systems updatedOnly where write access and approval policy allow it

7Outcome evaluatedAmendments, manifest survival at delivery, market challenges and post-release corrections

Amendments

Every amendment the lead makes is counted, including one made after the label shipped.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required8 items

Publishing a label, or the asset that carries it.

Deciding a work is evidently artistic under Art. 50(4).

Signing a provenance manifest for the organisation.

Holding editorial responsibility for a publication.

Automation boundaryAgent acts unaided

✓Assemble the manifest and record what each transform left.

✓Match each asset to the marking and disclosure tests it meets.

✓Re-check provenance at packaging, not only at ingest.

✓Flag what the evidence leaves ambiguous, and hold the file for the named owner.

Any write happens inside the boundaries agreed at implementation, never ahead of the approver.

Clearing an advertisement carrying a synthetic performer.

Telling an authority that other measures are adequate.

Accepting a vendor's marking claim as sufficient.

Changes to classification, marking or approval rules.

Example output

One asset label, annotated

Everything the agent assembles is attached to the evidence it was drawn from.

Disclosure output · single assetIllustrative example

Asset

Manifest finding

Transform

Source of record

Confidence

Disclosure form

Feature, EU release

Generative fill extended the background of a locked shot

Semantics altered

Vendor manifest on file

91%

Existence disclosure, end card

As receivedTaken from the pipeline record and the vendor's manifest — nothing on this side is written by the agent.

Evidence assembled Vendor manifest Transform log Delivery-package check

Why this formThe work is evidently artistic — the reduced duty a person still decides on.

ActionApproveAmendSend to legal review

What the score decidesBelow the configured threshold the file picks up a legal read before it reaches the lead.

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value landsValue 01 – 04

Every assetFrom the pipeline record

03Assembly

Build from what survived

Draw on the pipeline record, the vendor manifests and the per-market duties configured for the release.

01Approved path

A manifest signs an organisation

C2PA v2.4 makes the claim generator a non-human actor and the signer a certificate holder — a product and a company, not a person.

02Human review

Put the read where the duty bites

EU AI Act Article 50 applied on 2 August 2026, and the Digital Omnibus deferred the high-risk obligations rather than this one. Article 50(2) marking carries a grace period into December 2026 for generative systems already on the market.

04Build an evidence trail

The asset, the manifest it carries and the person who published stay on the record.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

Provenance and manifestsC2PA v2.4 manifests
CAWG identity assertions

Post and deliveryEditorial and VFX pipeline
Transcode and packaging

Media asset managementMAM and DAM systems
Metadata and sidecar stores

Agent

Content provenance & AI disclosure

Reads the asset
Assembles the file
Holds for approval

Publishing and compliancePlayout and streaming
Compliance case records

Observability & evaluationOpenTelemetry · Langfuse
Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

What sits between the model and a published label

Six controls wrapped around each other. What none of them catches is named in the map beneath.

L6 · Outermost — last line of defenceInward → L1 · closest to the model

L6Rollback / safe modeDrop to manifest reporting when evaluation or production signals degrade.Roll back

L5Version monitoringTrack model, prompt, classification-rule and market-configuration changes.Track

L4TraceabilityRecord the evidence, the classification, the amendments and the approval.Record

L3Approver sign-offHold the file for the named lead; once a label is published, no layer here retracts distributed copies.Gate

L2Policy guardrailsTest each file against the per-market marking and disclosure rules; a failure returns it unlabelled.Restrict

L1Confidence thresholdsRoute low-confidence classifications to a legal read first.Require review

Model coreClassification proposed — marking duty, disclosure form, market and confidence

L1 – L2Test whether a classification may stand

L3Puts the label in a named person's hands

L4 – L5Keep the asset and the manifest behind it

L6Drops to manifest reporting when signals degrade

How Nestack evaluates it

Evaluate the disclosure workflow — not only the published label.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the label the viewer sees

Depth of coverage ▼

E1Final-output evaluationDid the label match what the asset actually contains?

E2Step-level evaluationDid the agent use the pipeline record, the manifest and the market rules?

E3Tool evaluationDid it read the correct asset and write the correct market's file?

E4Confidence calibrationDo low-confidence classifications actually attract more amendments?

E5Slice evaluationHow does performance change across specific asset types?

E6Business outcomeHow many labels needed an amendment, or a correction after release?

Floor — the exposure the company carries

Failure modes

Where each failure originates in the agent

Seven failure modes, placed at the stage each one originates.

Agent lifecycleDirection of processing →

01 · Retrieval1 mode

WX-03

Ingest-only check

Provenance is read at ingest and never re-read.

Stage gathersPipeline record, manifests, transforms and duties

02 · Reasoning2 modes

WX-04

Artistic work over-read

A reduced duty is treated as no duty at all.

WX-06

Carve-out over-applied

A generative fill passes as standard editing.

Stage proposesMarking duty, disclosure form and confidence

03 · Tool / write2 modes

WX-02

File taken as label

An assembled file is treated as an approved label.

WX-05

Template assertion carried

A manifest asserts a history the asset lacks.

Stage writesOnly where write access and approval policy allow it

04 · Output1 mode

WX-01

Label not rendered

The placement chosen is not shown at first exposure.

Stage returnsThe label the viewer sees and the file behind it

05 · Change / Version1 mode

WX-07

Silent rule drift

A model or rule change narrows what is marked.

Stage tracksModel, prompt, classification and market rules

Sev-1 · a label ships outside the boundary Sev-2 · a wrong label reaches distribution Sev-3 · source degrades, asset goes to review

See the complete Media & Entertainment failure-mode catalogue →

Affected slices

Coverage in total can hide the market that fails

A slate-wide manifest-coverage figure can read as healthy while one market's assets carry most of the missing labels. Nestack reports the missing-label rate by slice, not only in total.

Slice performance — reported separately, not only in aggregateIllustrative example

Slice	Failure rate	Lift	Status
Vendor-supplied generative elements	8.7%	3.7×	Review
De-ageing and face replacement	6.1%	2.6×	Review
Restoration and upscaling passes	4.3%	1.8×	Watch
Camera-original masters	1.2%	0.5×	Normal

Bar: missing-label-rate lift vs. camera-original baseline · scale 0–4.0× · tick marks the 2.0× review threshold 2 of 4 slices over threshold

Evidence-linked improvement

An explanation does not close a cycle

A cycle is closed when the missing label is a case the next release must survive. That suite is what the next asset published is measured against.

Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn

01Detect

Missing-label rate rises in an asset slice.

02Diagnose

The trailer that went out unlabelled in one market is traced back to the packaging step that dropped its manifest.

03Improve

Number the correction, and the assets that surfaced it stay fastened to it.

04Verify

No release while a touched label case is outstanding; it reruns first.

05Learn

It becomes a fixture of the suite, and the marking rules follow it.

Learn → DetectThe return edge. The next detection runs against a suite one missing-label case longer.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, disclosure workflow, evaluation, integration, then production validation and handover.

Workstream Week 1Week 2Week 3Week 4Week 5Week 6

01Disclosure workflow discovery and boundary work.

02Pipeline and vendor source review.

03Market duty, carve-out and marking-rule mapping.

04Asset ingestion and manifest extraction.

05Transform tracking and evidence binding.

06Confidence scoring and flag routing.

07Compliance approval workflow.

08Publishing and delivery integration.

09Marking and disclosure cases.

10Guardrails and publication controls.

11Asset-trail instrumentation.

12Deployment, documentation and Agent Care handover.

12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel Final scope and sequence confirmed in discovery

Engagement tiers

What each tier includes

Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.

Capability✓ in scope · — not at this tier PilotOne market, one slate ProductionProduction delivery systems AdvancedMultiple markets / brands

Introduced at Pilot

Assembly to your pipeline and markets✓✓✓

Compliance approval✓✓✓

Manifest-coverage baseline✓✓✓

Introduced at Production

Reporting by market—✓✓

Approval workflow in your systems—✓✓

Approved write-back—✓✓

Production-pipeline integration—✓✓

Introduced at Advanced

Multi-market duty sets——✓

Multi-stage compliance approvals——✓

High asset volume——✓

Multi-market disclosure controls——✓

Build price From $5,000 From $8,000 Custom quote

Final build priceConfirmed after discovery based on integrations, workflow complexity, transaction volume, approval controls and deployment requirements.

Separate from buildBuild pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.

What we need from you

What you bring, and what we build with it

Each input maps to a piece of build scope and a week in the delivery timeline.

You bringWe build with it

01Your pipeline and the transforms an asset passes → Asset ingestion and manifest extractionWeek 1

02Assets you have already released with labels → Coverage baseline and transform-survival checksWeek 2

03Your vendor terms and your Code of Practice position → Market duty, carve-out and marking-rule mappingWeek 1

04Access to relevant APIs, feeds or exports → Pipeline and vendor source assessment, then integration setupWeek 2

05Labels you would not want challenged → Marking cases and the evaluation suiteWeek 4

06What no label may assert → Confidence scoring, flag routing, guardrails and publication controlsWeek 3

07Named approvers to publish the label → Compliance approval workflow, then pilot and production validationWeeks 5–6

Nothing else is required Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

Each phase covers the weeks it really occupies, so week 5 runs two things at once.

Phase W1W2W3W4W5W6

Discovery W1

Build W2 – W3

Evaluate W4 – W5

Pilot & Launch W5 – W6

Week focus W1Disclosure workflow discovery, duty mapping and the automation boundary W2Pipeline integration and the coverage baseline W3Assembly workflow, confidence logic and approval controls W4Evaluation suite, marking checks and failure-mode testing W5Delivery integration, pilot assets and targeted corrections W6One release slate run under the compliance lead, then Agent Care handover

Reading the bandEach bar covers only the weeks its work is named in. The week 5 overlap is real work, not padding.

At the end of W6Validation closes on live assets, and Agent Care assumes the monitoring.

DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Media & Entertainment AI agent

Build a provenance agent around the label a person signs.

Show us your pipeline, the markets you release into and who signs. Bring one release slate and we will map the transforms, the duties each market attaches and what stays with your compliance lead.

Discuss this agent →

Nestack Agents · Content provenance & AI disclosureAGT-ME-18 · Agent Care available after launch