Skip to content
AI Governance agents
Risk, Trust & ResilienceAI GovernanceDocumentation, Lineage & Evidence Readiness

AI Data & Dependency Lineage Agent

Maps data, models, prompts, retrieval sources and tools through downstream decisions.

Builds the traceable dependency chain from source records through transformations and retrieval to the decision or action a system takes. It highlights unowned datasets, stale indexes, hidden fourth-party models and circular dependencies that a flat inventory misses.

Authority

Inform

Team role

Provides specialist analysis

Handoffs

Named collaborators

The role

What it owns and where its authority ends

Desk

Documentation, Lineage & Evidence Readiness

Desk workflow

Inventory baseline, then the model or agent card, then data and dependency lineage, then the evidence-pack check, then owner attestation.

Collaboration

Works within a defined desk workflow

Decision boundary

Provides evidence or analysis without committing the decision.

Systems and capabilities involved

  • Data catalog

  • ML and prompt registries

  • Vector-store metadata

  • Software bill of materials

Handoffs

What this role gives and receives

Capabilities offered

Map AI data and dependency lineage

Return a source-to-decision graph with ownership, versions and confidence.

Receives:
System id, component manifests and runtime dependency evidence
Returns:
Lineage graph, material dependencies, stale edges and ownership gaps

External handoff

Data owners

External handoff

Enterprise architecture

External handoff

Third-party risk

Context

What the role needs to do the work

Current work
Changed component, candidate edges and unresolved ownership.
Prior interactions
Prior lineage snapshots, broken dependencies and remediation history.
Policies and reference
Data domains, system interfaces, model supply chain and decision taxonomy.
Working method
Lineage confidence, ownership and materiality rules.

Illustrative workflow

How the work moves

Starting point

Validation asks what evidence can influence a customer complaint response.

  1. 01

    Traverse runtime calls from response generation to retrieval, prompt and source systems.

  2. 02

    Resolve versions, owners, refresh clocks and external dependencies.

  3. 03

    Mark uncertain edges and generate a validation-ready dependency slice.

Result

A cited lineage graph with one stale index and one unowned external model dependency.

Checks and boundaries

What must be tested or reviewed

  1. 01Finds the external embedding service and source documents behind a vendor's managed retrieval product.
  2. 02Flags a production vector index whose refresh job stopped while the model card still claims daily updates.
  3. 03Does not infer an undocumented data flow solely from similar table names.

Human authority

  • Data owner confirms uncertain lineage
  • Architecture owner accepts external dependencies

Keep exploring