Creative AI Governance: Managing Production, Creation Records, and Ad Delivery
What you’ll learn
- The roles of OpenMontage, AIFX, and AdCP, and the scope each can manage
- Why project IDs, asset IDs, and approved versions need to remain consistent from production through delivery
- Which rights reviews, human approvals, and delivery conditions need to pass to the next owner
- Which mismatches and missing checks should stop an asset before delivery
Governance across an AI-assisted production workflow
If the file reviewed by the production team differs from the file ultimately delivered as an ad, unapproved content can go live even though an approval took place during production. AI-assisted production therefore needs more than separate checks at each stage. Each owner also needs to pass forward what was checked and which version was approved.
A Creative AI Governance reference architecture is a template for organizing those rules and review responsibilities. It connects production decisions, creation records for the finished work, and pre-delivery approval to the same project and version.
OpenMontage, AIFX, and AdCP provide useful examples. They are not products competing to solve the same problem. They address different scopes: production, creation records for finished work, and advertising delivery.
The sections that follow cover their roles and limits, the information that should pass between stages, where to begin based on an organization’s current problems, and what to verify before adoption.
The roles and limits of OpenMontage, AIFX, and AdCP
OpenMontage: production progress and approval records
OpenMontage divides production into stages such as research, proposal, script, scenes, assets, editing, and composition. It records drafts, human approvals, and the reasons for decisions at those stages. It also saves checkpoints—points from which interrupted work can resume—so the team does not need to start again from the beginning.[1]
Its primary users are production leads and reviewers responsible for individual stages. The workflow can help prevent skipped reviews and lost work after interruptions. OpenMontage alone, however, cannot determine whether a contract or consent remains valid or stop delivery through an external service.
The OpenMontage article explains this workflow in more detail.
AIFX and AIFI: creation records carried with finished work
AIFX, together with its AIFI image format, provides a way to package creation records with a finished work when it is transferred. Those records cover information about the creator, the creation context, the AI used, and cryptographic hashes that help detect whether a file has changed.[2]
The main participants are production staff who maintain the finished-work record, legal and rights reviewers, and the person who approves publication. Keeping the record with the finished work reduces the risk of separating the asset from its documentation or transferring a changed file without noticing. In practice, documentation of permitted uses should also remain linked to the finished work.
A creation record does not, by itself, establish the right to use the work. Identity verification, confirmation that the rights remain valid, and approval to publish are separate decisions.
The AIFX article examines this boundary in more detail.
AdCP: information exchange among advertising participants
AdCP is a protocol—a set of rules for exchanging information—used by the parties that prepare an ad proposal, review it against policies, and provide inventory and delivery.[3]
Relevant information includes the advertising request, creative preview, creation records, results of an independent review, policy-check results, the approved version, and delivery records. These shared structures help the orchestrator, independent reviewer, and delivery operator avoid confusing roles or versions.
An information-exchange mechanism alone does not guarantee that an ad rejected during review will be stopped by the delivery system. AdCP’s official documentation also explains that protocol seams do not themselves enforce a rejection or guarantee legal and policy compliance.[4]
A creation record supplied by the sender is not automatically information verified by an independent party. A separate reviewer still needs to assess the record, and the appropriate owner still needs to decide whether the rights remain valid.
See the AdCP article for further detail.
Comparison of the three roles
| System | Primary management scope | Decisions or actions still required elsewhere |
|---|---|---|
| OpenMontage | In-progress drafts, approvals, decision reasons, and restart points | Validity of contracts and consent; stopping delivery through an external service |
| AIFX and AIFI | Creation records attached to finished work and detection of file changes | Identity verification, validation of rights, and approval to publish |
| AdCP | Transfer of advertising requests, review results, approved versions, and delivery records | Independent review of creation records, validation of rights, and delivery stops based on rejection results |
This comparison is this article’s framework for evaluating enterprise use. It is not a shared classification officially defined by the three projects.
Shared management information from production through delivery
The purpose of the reference architecture is not to connect the three systems directly. The important capability is being able to verify, from production through delivery, whether the file currently being handled is the same work that was approved.
Three identifiers should therefore remain consistent across all stages:
- Project ID: identifies the project to which the work belongs
- Asset ID: identifies the specific work within that project
- Approved version: identifies the point-in-time version that received approval
If the version approved during production, the version packaged with creation records, and the version sent for delivery differ, unapproved work can be published even when every stage performed its own check.
Management scope at each stage
The reference architecture separates the path from production through delivery into three stages.
| Stage | What it manages |
|---|---|
| Production | The purpose of the work, AI generation, editing, quality review, human approval, and restart points |
| Storage of the finished work and its records | The finished work, version number, creation method, source-material locations, file hash, and rights-document locations |
| Pre-delivery review and delivery management | Destination requirements, brand and policy compliance, AI-use disclosure decisions, delivery status, and procedures for withdrawing a problematic ad |
Rules and decision records shared across all stages
Managing each stage separately still leaves room for gaps during handoffs. Alongside the shared identifiers, the organization should define who can view and change information, who approves it, which AI services are permitted, how sensitive the information is, and how long it is retained.
The same project and asset record should also preserve exceptions, reasons for changes, approval expiry dates, post-delivery issues, delivery stops, and withdrawal history.
This design does not require OpenMontage to become the production platform, every image to be converted to AIFI, or every publisher to support AdCP. The public material from these projects provides patterns for dividing responsibility and preserving records. Those patterns can also be applied to existing project management, finished-work storage, approval, and advertising delivery and tracking processes.
Handoff information and conditions that stop delivery
This article calls the agreement about who checks what and what information passes to the next stage the “Creative AI Governance Interface Contract.” Here, an interface contract means a handoff table that defines the information exchanged between stages and the responsibility for checking it. It is not a technical specification for automatically connecting products.
The table should identify the minimum required information, the accountable reviewer, and the conditions that stop a handoff when information is missing.
Handoff from production to the record owner
Production first records the relevant information, after which the owner of the finished-work record verifies and completes it. The English text in parentheses gives a field name that can be used in a system or management table.
| Shared field | Information recorded by production | Information verified or added by the record owner |
|---|---|---|
Project ID (campaign_id) | Project ID for the work | Project ID associated with the finished work |
Ad asset ID (creative_id) | Asset ID used consistently from planning onward | Asset ID linked to the complete record and document set |
Reviewed version (asset_version) | Version reviewed by a person | Version corresponding to the file hash |
Completed stage (workflow_checkpoint) | Completed stage and restart point | Reference to the corresponding creation record |
Creation-history reference (provenance_ref) | Location of the AI generation and editing history | Inventory of creation information and verification result |
Rights status (rights_status) | Source material and request for rights review | Reference to documents showing contracts, consent, and permitted-use dates |
Human approval (human_approval) | Approver, approval time, and approved version | Reference to the approval record |
Delivery policy (distribution_policy) | Intended use | Conditions for the purpose and delivery destination |
AI-use disclosure status (disclosure_status) | How AI was used | Records supporting the decision about whether to disclose AI use |
Decision record (audit_record) | Decisions and exceptions | Review and update history |
Creation history is also called provenance: information about where a work came from and how it was generated or edited. The decision record serves as an audit record for reconstructing what happened if a problem occurs.
Pre-delivery checks and accountable owners
The team then compares the work awaiting delivery with the information received in the handoff. Assigning an accountable owner to each check also establishes who must answer when information is missing.
| Shared field | What to verify before delivery | Accountable owner |
|---|---|---|
| Project ID | Matches the project for which inventory was purchased and delivery was requested | Overall coordinator |
| Ad asset ID | Matches the asset ID sent to the delivery provider | Production lead |
| Reviewed version | File awaiting delivery is the approved version | Production lead and delivery operator |
| Completed stage | No required stage remains incomplete | Production lead |
| Creation-history reference | Record and file can be checked at receipt, and the result is retained | Production lead and overall coordinator |
| Rights status | Intended use, territory, and destination fit the permitted scope | Legal and rights owner |
| Human approval | Approval is still valid, and neither the work nor its intended use changed afterward | Brand owner and business owner |
| Delivery policy | Matches the delivery provider’s rules | Delivery operator |
| AI-use disclosure status | If law or policy requires disclosure, its wording, placement, and implementation owner are clear | Brand owner and business owner |
| Decision record | The history of applying the approved version, delivery, stopping, and withdrawal can be recorded | Delivery operator |
These checks help prevent project and asset mix-ups, delivery of an unapproved version, skipped stages, and receipt of work whose creation history cannot be checked. They also help prevent use outside the permitted scope, reuse of expired approval or approval for a different work, delivery to an unintended use or destination, missing AI-use disclosure, and loss of the decision trail after a problem.
Conditions that prevent progress to the next stage
The handoff table needs stop conditions as well as review fields:
- Management IDs or file hashes do not match
- Rights review is incomplete
- Approval has expired
- No decision has been made about whether AI use must be disclosed to consumers
An asset meeting any of these conditions should not move automatically to the next stage. It should return to the owner accountable for the affected field. If the organization permits an exception, it should record the reason, approver, expiry date, and scope.
Where to begin based on the current problem
An organization does not need to improve production, records, and delivery at the same time. The right starting point depends on where the current problem occurs.
Frequent missed reviews or rework
The production lead maps the production process and identifies where a person must approve the work. Making clear who reviews each stage is the foundation for preventing required checks from being skipped.
Creation history and rights documents are stored separately
Production and legal or rights owners link one finished work to its creation method, permitted-use documents, review owner, and the stop procedure used when information is missing.
This collection can be treated as an AI-generated asset evidence packet: the asset plus the records supporting how it was made and the decisions about how it may be used. The packet prevents a finished work from reaching the next owner without its records.
The approved version differs from the delivered version
The delivery operator connects the approval record and delivery target with the same ad asset ID and version. This makes it possible to verify that the delivered file is the approved file.
Several problems occur in sequence
The overall coordinator puts the shared IDs, required decisions, and reasons to stop into one handoff table. This reduces the chance that the next owner will accept missing information without noticing it.
Scope of a small pilot
Limit a pilot to one advertising project, one type of work, and one delivery destination. This makes handoff problems easier to identify.
Production, legal and rights, brand, and ad-delivery owners use the same table to identify which information can be collected automatically, which decisions remain human responsibilities, and where required records are missing.
Pre-adoption checks and the limits of automation
A useful design reference is not necessarily ready to serve as the primary product for day-to-day work. Before adopting OpenMontage, AIFX, or AdCP, evaluate the working features and connections to existing processes as well as the ideas described in their documentation.
Checks for capability, connection, and operation
| Area | What to verify |
|---|---|
| Working capability | The required features and information destinations exist in practice, not only in descriptions |
| Connection to other tools and long-term storage | Old and new versions and different tools can read the same record; old records can be migrated and opened after long-term storage |
| Rejection and expiry handling | The delivery side reliably stops an item after policy rejection or expiry |
| Information protection | Prompts, personal information, and unpublished material are not copied more widely than necessary |
| Terms and operating model | Terms, operating ownership, incident responsibility, and record export on exit fit the organization’s requirements |
Checks suited to automation and decisions left to people
Trying to automate every decision can allow an item to pass merely because a field is nonempty, even when the context of rights and use has not been adequately reviewed.
Automation is best centered on matching management IDs, checking file formats, detecting file changes, checking dates, and moving information between stages. Rights, brand compliance, and exceptions remain decisions for their accountable human owners.
Summary: consistent management of approved work and decision records
OpenMontage, AIFX, and AdCP are useful references for thinking about production, creation records for finished work, and advertising delivery. When using their patterns together, the shared foundation is the project and ad asset IDs, approved version, rights status, AI-use disclosure decision, and decision record.
Passing that information with the work lets the next owner distinguish what has been checked from what remains unverified. A creation record alone does not establish that rights review and publication approval are complete.
When evaluating this design, test whether the organization can stop work with missing records, return it to the appropriate owner, and resume under the same management ID after correction. Following one advertising project with a single handoff table can test whether this operating model works before adopting any of the individual tools.
This article is a general information summary and is not legal advice. Confirm practical decisions with a qualified specialist.
References
- OpenMontage, README
- AI First Exchange, AI-First Exchange
- Ad Context Protocol, AdCP - Advertising Context Protocol
- Ad Context Protocol, Trust & Security
Check the official sites and documentation for the latest release and update details.