To calculate the AI trust score, Collibra evaluates the following factors for each supported asset type. Each factor produces a score between 0 and 100. The overall trust score is the weighted average of all active factor scores: each factor score is multiplied by its configured weight, the results are summed, and the total is divided by the sum of all weights. Factors with a weight of 0 are excluded from the calculation entirely.
The trust score updates automatically, typically within seconds, whenever relevant data on the asset changes, for example when you link a related asset, advance the lifecycle, or complete an assessment.
For information on configuring factor weights and score ranges, go to Edit the AI Trust Score settings. For information on interpreting and improving scores, go to Interpreting and improving the AI trust score of an asset.
Supported asset types
The AI Trust Score is calculated for the following asset types. The score calculation differs depending on the asset type.
| Asset type | Public ID | UUID | Trust score type |
|---|---|---|---|
| AI Use Case | AIUseCase | 00000000-0000-0000-0000-000000031401 | AI Use Case trust score |
| AI Agent | AIAgent | 00000000-0000-0000-0000-000000031450 | AI technical asset trust score |
| AI Agent Version | AIAgentVersion | 00000000-0000-0000-0000-000000031452 | AI technical asset trust score |
| AI Model Version | AIModel | 00000000-0000-0000-0000-000000031402 | AI technical asset trust score |
| AI Base Model | FoundationalAIModel | 00000000-0000-0000-0000-000000031422 | AI technical asset trust score |
Factor 1: Documentation
Applies to all supported asset types. This factor checks how well the asset is described and connected in the catalog, using three yes/no checks.
Check 1: Has an Owner assigned
The asset must have at least one user or group assigned with the Owner role.
| Check | Role (Public ID) | UUID | Role in Trust Score (theme) |
|---|---|---|---|
| Has an Owner assigned | Owner | 00000000-0000-0000-0000-000000005040 |
Documentation Applies to all supported asset types. The asset must have at least one user or group assigned with the Owner role. |
Check 2: Has a description
The asset must have a non-empty description attribute. For technical assets, a description imported from a source system also counts.
| Attribute type | Public ID | UUID | Notes |
|---|---|---|---|
| Description | Description | 00000000-0000-0000-0000-000000003114 | All asset types |
| Description from source system | DescriptionFromSourceSystem | 00000000-0000-0000-0001-000500000074 | Technical assets only |
Note Collibra looks for a Description or a Description from source system. If either or both are present, the criterion is considered met. Having both does not carry more weight than having only one. For AI Use Case assets, only the standard Description attribute is checked; Description from source system is not used.
Check 3: Is connected to related assets
The direction of this check differs by asset type.
| Asset type | What is checked | Public ID | UUID |
|---|---|---|---|
| AI Use Case | Is linked to at least one technical AI asset (AI Model Version, AI Agent, or AI Base Model) |
|
|
| AI Model Version | Is used by at least one AI Use Case (directly or via its parent Base Model) |
|
|
| AI Base Model | Is used by at least one AI Use Case (directly or via one of its Model Versions) |
|
|
| AI Agent | Is used by at least one AI Use Case | AIUseCaseUsesAIAgent (as target) | 00000000-0000-0000-0000-000000007109 |
| AI Agent Version | The parent Agent is used by at least one AI Use Case | AgentHasVersion ← AIUseCaseUsesAIAgent | 00000000-0000-0000-0000-000000007206 ← 00000000-0000-0000-0000-000000007109 |
The Documentation score is derived from the number of checks passed:
| Checks passed | Score |
|---|---|
| 3 of 3 | 100 |
| 2 of 3 | 60 |
| 1 of 3 | 40 |
| 0 of 3 | 0 |
Factor 2: Lifecycle progress
Applies to all supported asset types. This factor measures how far the asset has progressed through its lifecycle.
If Lifecycle management is disabled for an asset type, the weight for this factor is set to 0 and it is excluded from the score entirely. Assets that transition into the Retirement lifecycle phase have their trust score cleared and no longer receive a score.
The score increases incrementally as the asset advances through its lifecycle statuses. For example, in a lifecycle with 4 statuses, an asset at status 2 scores 50.
Factor 3: Risk rating
Applies to AI Use Case assets only. This factor checks the Overall Risk Rating attribute set on the AI Use Case. It reflects whether the risk has been assessed and what the outcome was.
The value is the answer to the "What is the overall risk rating of this AI use case?" question in the out-of-the-box Risks and Safeguards assessment.
| Attribute type | Public ID | UUID |
|---|---|---|
| Overall Risk Rating | OverallRiskRating | 00000000-0000-0000-0000-000000000325 |
| Overall Risk Rating value | Score |
|---|---|
| Low | 100 |
| Medium | 50 |
| High | 10 |
| Not set | 0 |
Note A High risk rating scores 10 rather than 0, acknowledging that a risk assessment was completed even if the outcome is unfavorable.
Factor 4: Data Assets
Applies to AI Use Case assets only. This factor checks whether the AI Use Case is linked to at least one data asset. Any single link of the types listed below is sufficient.
| Relation type | Public ID | UUID |
|---|---|---|
| AI Use Case infers from / used to infer Asset | AIUseCaseTransformsAsset | 00000000-0000-0000-0000-000000007099 |
| AI Use Case is trained by / trains Asset | AIUseCaseTrainedByAsset | 00000000-0000-0000-0000-000000007100 |
| AI Use Case has output / is output Asset | AIUseCaseHasOutputAsset | 00000000-0000-0000-0000-000000007101 |
| Condition | Score |
|---|---|
| At least one data asset linked | 100 |
| No data assets linked | 0 |
Factor 5: Linked Technology Assets
Applies to AI Use Case assets only. This factor reflects the average trust score of the technical AI assets linked to the Use Case.
| Relation path | Public ID | UUID |
|---|---|---|
| AI Use Case → AI Model Version | AIUseCaseUsesAIModel | 00000000-0000-0000-0000-000000007098 |
| AI Use Case → AI Agent | AIUseCaseUsesAIAgent | 00000000-0000-0000-0000-000000007109 |
| AI Use Case → AI Base Model | AIUseCaseUsesBaseModel | 00000000-0000-0000-0000-000000077002 |
| AI Use Case → AI Base Model → AI Model Version | AIUseCaseUsesBaseModel → ModelHasVersion | 00000000-0000-0000-0000-000000077002 → 00000000-0000-0000-0000-000000077000 |
| AI Use Case → AI Agent → AI Agent Version | AIUseCaseUsesAIAgent → AgentHasVersion | 00000000-0000-0000-0000-000000007109 → 00000000-0000-0000-0000-000000007206 |
Note If the same asset is reachable via multiple relation paths, it is counted only once in the average. Updating any linked technical asset's trust score automatically triggers recalculation of the AI Use Case score.
The score is the average trust score of all linked technical assets in scope, rounded up. If no linked technical assets are in scope, this factor scores 0.
Factor 6: Data integrity
Applies to AI Agent, AI Agent Version, AI Base Model, and AI Model Version assets. This factor evaluates how well the data used by the AI asset is governed, using two independent signals: lineage and sensitivity.
Lineage checks whether the asset is linked to data assets (training data, inference data, or output data). Sensitivity checks whether any of those linked data assets are classified with a sensitive data category (PII or Personally Identifiable Information). The evaluation logic differs by asset type.
AI Model Version
Lineage: Checks the model version directly for any of the following relation types.
| Relation type | Public ID | UUID |
|---|---|---|
| AI Model Version trained by / trains Asset | AIModelTrainedByAsset | 00000000-0000-0000-0000-000000007102 |
| AI Model Version infers from / used to infer Asset | AIModelInfersFromAsset | 00000000-0000-0000-0000-000000007103 |
| AI Model Version has output / is output Asset | AIModelHasOutputAsset | 00000000-0000-0000-0000-000000007104 |
Sensitivity: Checks data categories reachable via the following derived relation type paths for PII classification.
| Derived relation type (Public ID) | Meaning |
|---|---|
| AIModelVersionInfersFromDataCategory | Data category of the inference data used by the model version |
| AIModelVersionIsTrainedByDataCategory | Data category of the training data used by the model version |
| AIModelVersionHasOutputDataCategory | Data category of the output data produced by the model version |
AI Base Model
Lineage: Checks the base model directly for the following relation type.
| Relation type | Public ID | UUID |
|---|---|---|
| AI Base Model trained by (AI Governance) / trains (AI Governance) Asset | FoundationalAIModelTrainedByAsset | 00000000-0000-0000-0000-000000007117 |
Sensitivity: Checks data categories reachable via the following derived relation type paths for PII classification.
| Derived relation type (Public ID) | Meaning |
|---|---|
| AIBaseModelInfersFromDataCategory | Data category of the inference data used by the base model |
| AIBaseModelIsTrainedByDataCategory | Data category of the training data used by the base model |
| AIBaseModelHasOutputDataCategory | Data category of the output data produced by the base model |
AI Agent
Lineage is evaluated via two paths: model lineage (delegating to the AI Model Versions linked to the agent) and tool-based data source connections. Sensitivity is evaluated only via linked AI Model Versions, using the same checks as for AI Model Version.
| Relation type | Public ID | UUID |
|---|---|---|
| AI Agent uses / is used by AI Model Version | AIAgentUsesAIModel | 00000000-0000-0000-0000-000000007108 |
| AI Agent Tool can consult data source / can be consulted via Inference Data | AIAgentToolCanReadInferenceData | 00000000-0000-0000-0000-000000007211 |
| AI Agent can manipulate data in / can be manipulated via Output Data | AIAgentCanManipulateDataSource | 00000000-0000-00fc-0000-000000000025 |
AI Agent Version
Lineage is evaluated using two independent sources: tool lineage and parent agent model lineage.
Tool lineage: Finds tools called by this Agent Version, then checks those tools for data links.
| Relation type | Public ID | UUID |
|---|---|---|
| AI Agent Version calls / is called by Tool | AgentVersionCallsTool | 00000000-0000-0000-0000-000000007207 |
| AI Agent Tool can consult data source / can be consulted via Inference Data | AIAgentToolCanReadInferenceData | 00000000-0000-0000-0000-000000007211 |
| AI Agent Tool can manipulate data in / can be manipulated via Output Data | AIAgentToolCanManipulateOutputData | 00000000-0000-0000-0000-000000007212 |
Parent agent model lineage: Follows AgentHasVersion (00000000-0000-0000-0000-000000007206) to the parent AI Agent, then applies the same lineage evaluation as AI Agent: model version connections (via AIAgentUsesAIModel) and tool-based data source connections (via AIAgentToolCanReadInferenceData and AIAgentCanManipulateDataSource).
Note Sensitivity is only evaluated from the parent AI Agent's model versions (same PII check as AI Model Version). Tools are not checked for sensitivity.
The Data integrity score is derived from the combination of lineage and sensitivity signals:
| Lineage | Sensitivity | Score |
|---|---|---|
| Yes | Yes | 100 |
| Yes | No | 60 |
| No | Yes | 60 |
| No | No | 0 |
Factor 7: Linked versions score
Applies to AI Base Model and AI Agent assets. This factor reflects the average trust score of the version assets linked to the asset:
- AI Model Version assets for AI Base Model assets
- AI Agent Version assets for AI Agent assets.
The default weight for this factor is 1 for both asset types.
| Relation type | Public ID | UUID |
|---|---|---|
| AI Base Model has version / is version of AI Model Version | ModelHasVersion | 00000000-0000-0000-0000-000000077000 |
| AI Agent has version / is version of AI Agent Version | AgentHasVersion | 00000000-0000-0000-0000-000000007206 |
Note If an asset has no linked versions, this theme is excluded from the trust score and does not affect the score. As a result, this theme may appear on some AI Agent and AI Base Model assets but not others in the same instance. This is expected behavior.
The score is the average trust score of all linked version assets in scope, rounded up. If no linked version assets are in scope, this factor is excluded from the score entirely.
Factor 8: Operational health
Applies to AI Agent Version assets only. This factor measures the operational health of the AI Agent Version based on its recent monitoring pass rate: the percentage of monitoring checks that passed over the last 7 calendar days. It requires the AI Monitor flow feature to be enabled in the Collibra configuration. The default weight for this factor is 1. For details on how the pass rate is calculated and how AI monitoring works, go to AI monitoring for AI agents.
| Situation | Score |
|---|---|
| No monitoring data available for this agent for the past 7 days | Excluded — weight set to 0, not counted in score |
| AI Monitor flow is disabled in configuration | Excluded — weight set to 0, not counted in score |
| Monitoring data is available | 7-day average monitor pass rate (0–100) |