AI Control Library / CF-01: Lifecycle Governance
Designation: Core | Control types represented: Operational and Technical
A model version alone does not identify the deployed system. Prompts, retrieval indexes, filters, tools and permissions can change its behavior even when model weights remain the same.
Control requirement
Establish and maintain an approved baseline for each consequential AI deployment, covering the components and settings needed to explain, reproduce where feasible, assess and restore its behavior. Record changes and externally managed uncertainties.
Why this matters
The control is effective when its required decisions and safeguards can be demonstrated in the actual workflow. An approved document or completed form alone is insufficient if the described actions are not performed.
How the control types work together
Operational controls define the process and decision authority; Technical controls enforce boundaries or produce the system evidence needed to verify it.
When it applies
Deployed AI services and their development or recovery environments where differences can affect assurance or restoration.
Tailor depth and review frequency to potential impact, system authority and uncertainty. Document the reason for any exclusion; a supplemental designation does not make an otherwise applicable safeguard optional.
Ownership
The system owner is accountable; configuration and platform teams maintain baseline records and authorized repositories.
Implementation guidance
- Identify model and provider versions, prompts, data and index snapshots, filters, runtime settings, dependencies, tools and access scopes.
- Record approved values, provenance, integrity references and environmental dependencies in a versioned baseline manifest.
- Link every deployment and approved change to its baseline; protect manifests and configuration from unauthorized modification.
- Record hosted-service elements that cannot be pinned or observed, and define change detection, evaluation or containment for that uncertainty.
How to verify it
Use a representative sample of actual records or an authorized controlled exercise. A reviewer with suitable expertise should compare observed practice with the approved scope and acceptance criteria.
| Test | Expected result |
|---|---|
| Trace a live deployment | A complete approved manifest identifies the behaviorally relevant components. |
| Reconstruct the baseline in a controlled environment | Required artifacts are available or documented provider limitations have an accepted response. |
| Compare runtime settings with the manifest | Material differences are detected and investigated. |
Record a finding when a required result is absent, evidence cannot substantiate the claim, or practice differs materially from the approved conditions. An exception may record accepted risk, but does not turn a failed safeguard into an effective one.
Evidence to retain
- Baseline manifest and approval
- Artifact and configuration references
- Deployment-to-baseline links
- Provider limitations and reconciliation results
Protect evidence according to its sensitivity and approved retention requirements. Retain enough detail to substantiate the assessment without unnecessarily duplicating confidential or personal information.
If the control fails
Escalate an unknown or materially inconsistent baseline to the system and release owners. Restrict consequential use when the organization cannot establish what is running or whether approval conditions still hold.
Example in practice
Hypothetical example: Two assistants use the same model but different retrieval indexes and tool permissions. Separate manifests make the differences visible and prevent an evaluation of one deployment being misapplied to the other.
Related controls
- CF01-009: AI Change Management Policy & Procedures
- CF01-S03: AI Baseline Deviation Detection
- CF01-S11: AI System Backup
Framework mappings
- CSA AICM (CCC-06)
- NIST AI RMF (MAP 1.1)
- EU AI Act (Art. 9(7))
- ISO 42001 (Cl. 8.2)
Review and maintenance
Review at the interval approved by the control owner and after a material change, relevant incident or failed assessment. Reassess the affected scope, update records and verify corrective actions before closing findings.
Version 1.0