AI Control Library / CF-01: Lifecycle Governance
Designation: Core | Control types represented: Operational
A prompt edit, retrieval update or new tool permission can alter behavior as much as a model replacement. A software-only change process can miss the changes that expand AI authority or change its consequences.
Control requirement
Control changes to AI systems through documented procedures that classify impact, require proportionate review and testing, establish authorization, and address rollback and post-change observation. Include changes beyond model code.
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 provide the policies, decisions and recurring processes described here.
When it applies
Models, prompts, datasets, retrieval content, configurations, permissions, tools, infrastructure and provider-driven changes affecting deployed behavior.
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 change-management owner maintains the process; system owners classify changes and release authorities approve implementation.
Implementation guidance
- Define change categories and materiality criteria, including autonomy, data access, affected populations and provider updates.
- Require a change record with purpose, affected baseline, impact analysis, dependencies, test scope and rollback approach.
- Establish review and approval gates proportionate to risk, with a controlled emergency route and traceable implementation records.
- Observe the system after release and reconcile actual changes against the approved record; route deviations and unexpected impacts for response.
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 |
|---|---|
| Sample recent prompt, permission and vendor changes | Each entered the appropriate review path. |
| Trace a high-impact release | Testing, authorization and rollback readiness preceded implementation. |
| Simulate an emergency change | Expedited approval and subsequent review remain documented. |
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
- Change policy and classification criteria
- Change records and approvals
- Deployment and baseline records
- Post-change review
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 unapproved or misclassified changes to the release authority. Freeze further expansion and assess rollback or containment when behavior or permissions exceed the approved scope.
Example in practice
Hypothetical example: A retrieval assistant gains an email tool. The change process treats the new external-action capability as material, requiring authorization and tool-boundary tests instead of approving it as a simple integration update.
Related controls
- CF01-010: AI Change Quality Testing & Approval
- CF01-011: AI Change Authorization
- CF01-013: AI System Rollback & Restoration
Framework mappings
- CSA AICM (CCC-01)
- NIST AI RMF (MANAGE 4.2)
- EU AI Act (Art. 9(2)(d))
- 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