Is an Asset Change Material?Enterprise
A field-by-field lookup of which AI asset changes count as material for use case versioning, and which are cosmetic.
When you change an AI asset that a use case governs, Airia classifies the change as material or cosmetic. This page lists the verdict for each field. For what a material change does to a use case, see Use Case Versions.
Every change is material except a rename. A field that is not listed on this page is also treated as material, because an unrecognized change is never assumed to be safe.
| Verdict | What it means for a use case |
|---|
| Material | Counts as a substantive change. For a use case live in Monitor, it cuts a new version and returns to Assess (unless automatic version cutting is off). At Mitigate or Approve, its Nutrition Label is refreshed in place. Earlier than that, its risk analysis re-runs. |
| Cosmetic | Recorded and shown on the Nutrition Label, but never sends a use case back, refreshes its label, or re-runs its analysis. |
| Not tracked | Not part of the asset's governed definition, so no asset version is recorded. |
Agents discovered from a connected provider, such as Copilot Studio, Bedrock, or Vertex.
| Field | Verdict |
|---|
| Name | Cosmetic |
| Agent kind | Material |
| Instructions (system prompt) | Material |
| Autonomous (on/off) | Material |
| Number of autonomous triggers | Material |
| Generative AI enabled | Material |
| Agent-to-agent enabled | Material |
| Web browsing capability | Material |
| Authentication mode | Material |
| Require users to sign in | Material |
| Accessible org-wide | Material |
| Publicly accessible | Material |
| Access control policy | Material |
| Active / inactive | Material |
| Guardrail policy (added, removed, or swapped) | Material |
| Guardrail policy name, mode, base policy, or content filters | Material |
| Models | Material |
| Tools | Material |
| MCP servers | Material |
| Knowledge sources | Material |
Changes are tracked step by step. A step is identified by its title, so renaming a step counts as removing one step and adding another, which is material.
| Field | Verdict |
|---|
| Agent name | Cosmetic |
| Step added or removed | Material |
| Step renamed | Material |
| Step type | Material |
| Step configuration | Material |
| Step guardrail | Material |
| What the step invokes | Material |
| Step model | Material |
| Step prompt (any part of it) | Material |
| Step Python code | Material |
| Step data sources | Material |
| Step tools | Material |
| Step MCP servers | Material |
| Field | Verdict |
|---|
| Name | Cosmetic |
| AI models | Material |
| AI features in scope | Material |
| Provider | Material |
| Category | Material |
| Risk level | Material |
| Risk flags | Material |
| Compliance certifications | Material |
| Sub-processors | Material |
| Deployment options | Material |
| Data sent to AI | Material |
| Data retention | Material |
| Encryption | Material |
| Training on customer data | Material |
| Field | Verdict |
|---|
| Name | Cosmetic |
| Provider | Material |
| Is an AI service | Material |
| OAuth scopes | Material |
| Field | Verdict |
|---|
| Name | Cosmetic |
| Vendors | Material |
| Categories | Material |
| LLM API base URLs | Material |
| MCP servers | Material |
| Minimum installed version | Material |
| Field | Verdict |
|---|
| Name | Cosmetic |
| Transports | Material |
| Endpoints | Material |
| Authentication types | Material |
| Scopes | Material |
| Replaceable tier | Material |
| Sanctioned / unsanctioned | Material |
| Has a plaintext secret | Material |
| Tool names | Material |
| Change | Verdict |
|---|
| Asset deleted at the source | Material (recorded as a deletion) |
| A field not listed on this page | Material (the safe default) |
| A rename together with any other change | Material. One material field makes the whole change material. |
| First time the asset is seen | Neither. This is the starting point, not a change. |
| Security score, usage counters, last-refreshed time, raw provider data, provider version number | Not tracked |
