AI System Life Cycle
Description
Addresses AI-specific lifecycle management including design principles, development practices, testing and validation, change control, and retirement procedures.
Controls
| ID | Control | Description |
|---|---|---|
| A.5.1 | AI system design | Systematic design process incorporating safety, security, fairness, transparency, and accountability by design from ince... |
| A.5.2 | AI system development | Disciplined development practices including version control, peer review, documentation, and responsible AI principle ad... |
| A.5.3 | AI system verification and validation | Rigorous testing of AI systems for accuracy, fairness, robustness, security, and compliance before and after deployment. |
| A.5.4 | AI system deployment | Controlled deployment with phased rollout, monitoring, human oversight activation, and documented approval from accounta... |
| A.5.5 | AI system change management | Managed changes to AI systems including model updates, data changes, and configuration modifications with impact assessm... |
| A.5.6 | AI system retirement | Planned retirement or decommissioning of AI systems with data retention, transfer procedures, and stakeholder communicat... |
Implementation Guidance
Adopt AI design patterns embedding fairness constraints, privacy preservation, and explainability mechanisms. Implement MLOps pipelines with automated testing for model drift, bias, and adversarial robustness. Require multi-stage deployment (canary, blue-green) for high-risk systems with defined rollback criteria. Establish change advisory board review for significant AI system modifications with mandatory regression testing.
Evidence Requirements
AI system design specifications
Development standards and code reviews
Test plans and validation reports
Deployment approvals and monitoring logs
Change requests and impact assessments
Retirement procedures and records