Add an AI System

In the actual product, adding a system is not a single blind form submit. The drawer follows a real two-phase workflow: first classify, then save. That split is important because the second step expands or tightens the fields shown depending on the classification result and the legal pathway involved.

Step 1 — Describe The System Before Classification

The first screen of the real registration drawer asks only for the fields needed to produce a defensible first classification. This avoids overwhelming the user too early while still forcing enough context to distinguish a generic AI feature from a genuinely regulated use case.

Step 2 — Review Classification And Complete The Record

After step 1, ActLoom classifies the system and unlocks the save form. In the real UI, this second stage is visually disabled until classification exists. That design matters: the platform wants legal characterization to happen before the full record is stored.

Conditional Fields That Appear Only When Relevant

The real registration page does not show every regulatory field to every user. It reveals additional blocks only when a system enters a pathway where those details matter. This keeps the flow usable for minimal-risk systems while still supporting high-risk, product-safety, Annex X, and GPAI cases.

What The Classification Engine Returns

In the live product, classification is not just a label. The engine returns a package of decision support that the user can review before saving the record: risk category, suggested roles, Annex III signal, confidence level, and legal reasoning.

• Risk category, including prohibited, high-risk, limited-risk, minimal-risk, GPAI, or systemic GPAI.
• Confidence score so the team can judge whether a case is straightforward or needs more review.
• Legal reasoning in plain language to explain why the system landed in that category.
• Suggested role mapping such as provider or deployer, based on how the system is built and used.
• Annex III domain hint when the use case appears to fall in a high-risk sector.

Validation Rules The Registration Flow Enforces

The backend does more than store what the user typed. It blocks combinations that are inconsistent with the product’s current EU AI Act logic, which means the registration page also acts as a compliance quality gate.

Risk Classification

The registration drawer ultimately maps each system into one of the platform’s risk pathways. Those pathways are not just cosmetic badges: they change which tabs, documents, assessments, and validations become relevant later.

What Happens After You Save A System

In ActLoom, saving the system is the beginning of the compliance lifecycle, not the end of a form. The product uses the new record immediately to initialize downstream work.

System Inventory

Once saved, the system appears in the shared inventory. The inventory is not just a list: it is the operational map of what AI exists in the company, how risky each system is, how mature its compliance work is, and where owners should focus.

Keeping The System Record Accurate Over Time

AI systems change. The model version changes, the business use expands, a deployer becomes a provider through modification, or a limited-risk use case becomes high-risk once deployed in a sensitive domain. The registration record must keep up with those changes.