| 🏠 Back to Exam Syllabus | 📺 RooCloud on YouTube | 🌐 RooCloud Practice Exams |
AI Life Cycle Phases: From Plan and Design to Decommission
This episode of the ISACA Advanced in AI Security Management (AAISM) exam prep series walks through the full life of an AI solution and the risks that surface at every stage. It frames the life cycle as a familiar parallel to traditional system development, but with the added complexity that risk shifts shape from planning through retirement, and that everyone involved shares responsibility for an outcome that is trustworthy and fit for purpose.
What this episode covers
- How the AI life cycle parallels the system development life cycle and why risk shifts shape across phases.
- The plan and design phase as the blueprint for goals, stakeholders, initial risk assessment, and data governance.
- The collect and process data phase with its emphasis on consent, quality, completeness, and provenance.
- The build or adapt phase and why explainability and human-in-the-loop oversight are non-negotiable design choices.
- The test, evaluate, verify, and validate phase, including AI red teaming and use of the risk register.
- The deploy phase covering pilots, legacy compatibility, documentation as compliance evidence, and the people side of rollout.
- The operate and monitor phase for intended and unintended consequences, plus a clean retire or decommission plan.
Watch the full episode above for the worked examples and detailed explanations of each concept.
Frequently Asked Questions
What are the seven phases of the AI life cycle?
The AI life cycle moves through seven phases: plan and design, collect and process data, build or adapt models, test and validate, make available for use or deploy, operate and monitor, and retire or decommission. Each carries its own risk, from bias in design and data, through explainability and red teaming in build and test, to compatibility in deployment and obsolescence at retirement.
Why is the plan and design phase so important?
Plan and design is where you articulate the system’s concept, objectives, assumptions, and requirements, define clear goals and success metrics, document the business use case, and engage stakeholders early. It is also where you run an initial risk assessment, hunt for bias in data or algorithms, plan mitigations, and lay a foundation for data governance. Done well, this phase becomes the blueprint for the entire life cycle.
What is the difference between verification and validation in AI?
Verification checks the model was built correctly to specification, while validation checks it actually serves its intended purpose with accurate, reliable output. The phase uses several techniques including model testing against benchmarks like accuracy and recall, stress testing with extreme cases, comparative analysis against baselines, bias and fairness checks, and scenario analysis.
What is AI red teaming?
AI red teaming is a standout practice in the test and validate phase where you emulate real adversaries to find blind spots and weaknesses, both in the base model and at the application level. Whatever the testing reveals should drive fixes before production, with any remaining risk documented in the risk register and the solution authorized for release through change management.
When should an AI system be decommissioned?
Retirement usually happens when a solution becomes obsolete, a better alternative emerges, or ongoing risk assessment finds the risk now exceeds appetite. Once the decision is made, create a detailed decommissioning plan through change management covering data migration or deletion, system disentanglement, and continuity of operations, with clear communication to stakeholders to minimize disruption.
📚 Master the ISACA AAISM Exam!
Ready to test your knowledge? Access chapter-specific Multiple Choice Questions (MCQs) and full-length practice exams for the ISACA AAISM certification at RooCloud.com. Solve the chapter-wise questions to reinforce this lesson before moving to the next episode.
Reference: This article is based on concepts discussed in AI Life Cycle Phases: From Plan & Design to Decommission.