AI Act Register: Put the Scope Gate Before the Model

The current consolidated AI Act ties obligations to system purpose, role and context. An inventory therefore needs an owned record before it needs generated prose.

Inventory before interpretation

Each system record should identify:

• business and technical owners;
• purpose and operating context;
• provider and deployment arrangement;
• data categories and users;
• system status and approved role;
• source evidence and its freshness;
• open gap, accountable person and review date.

The person responsible for the record supplies or confirms the facts. A model can help organize material after those inputs are present.

Put a deterministic gate before drafting

The workflow has a simple order:

1. an operator confirms the system role and current status;
2. a rule selects the approved article set or scope values;
3. the model drafts reasons and missing-evidence notes in a fixed schema;
4. validation checks types, allowed values and record references;
5. a reviewer accepts, edits or returns the entry;
6. the register stores the evidence and decision.

The model receives a constrained task. It does not choose the legal scope or write directly to the authoritative register.

Record fields that support ownership

Keep unsupported values visible. A blank field is a review task; a guessed value is a false record.

Triggers and review ownership

Open a review when the company buys a new tool, changes a model or data source, expands users, changes purpose or receives new guidance. The trigger should create a task for a named owner and preserve the previous record version.

The register should expose stale evidence, unowned gaps and entries awaiting specialist review. A dashboard of completed drafts is less useful than a queue the owner can act on.

Test the gate as a product

Test role and status combinations, invalid values, missing evidence, changed records and model output outside the approved schema. Include an export check and a history check so a reviewer can explain why an entry has its current state.

The AI Act register case describes this pattern in a controlled workbench. Its case evidence is a reference for architecture and review flow; each organization still needs its own inventory and specialist assessment.

Decide whether this pattern fits

Use a register gate when the organization has multiple systems, recurring changes and a need to connect obligations to evidence. A spreadsheet can be enough for a small, stable inventory. The purchase decision should follow the operating burden the team can maintain and the process benefit model drafting provides.

Discuss an AI inventory workflow with Syntalith if you want to map records, owners, triggers and the first controlled scope.