AI system lifecycle
Stages of AI development from data to monitoring.
How the laws define this
Consensus across regulatory instruments establishes that the AI system lifecycle encompasses the full sequence of stages from initial conception, design, or research through data processing, model development, testing, deployment, operation, and eventual retirement or decommissioning. Frameworks uniformly apply this lifecycle structure to organize risk assessments, governance, safety controls, documentation, and ongoing monitoring throughout a system's existence. Divergence arises primarily in the specific operational phases explicitly bound to the lifecycle. For example, the Principles for the Ethical Use of Artificial Intelligence in the United Nations System uniquely incorporates initial procurement as a formal stage, and ISO/IEC 23894:2023 - Information technology — Artificial intelligence — Guidance on risk management explicitly includes operational incident handling. In contrast, instruments like the OECD Recommendation on Artificial Intelligence (AI Principles) and the Interim Generative AI guidance for the public service outline broader high-level phases without detailing procurement or incident handling.
Synthesised from the 8 statutory definitions below. Each one is quoted in full, with its source.
How jurisdictions differ
What is distinctive in each definition, binding law in force first. Showing 5 of the 8 instruments that define this term; all are listed under Definitions below.
| Jurisdiction | Instrument | Status | What is distinctive |
|---|---|---|---|
| OECD | Recommendation of the Council on Artificial Intelligence | In Force (Amended) | Broadly encompasses design and development, deployment, operation and monitoring, and retirement/decommissioning, with activities to manage the system across its operational life. |
| ISO | ISO/IEC 23894:2023 - Information technology — Artificial intelligence — Guidance on risk management | In Force | Emphasizes risk management by detailing progression from conception through data acquisition, model development, verification, validation, operational monitoring, incident handling, and retirement. |
| Switzerland | Strategy: Use of AI systems in the Federal Administration | In Force | Specifies a structured sequence of seven stages—plan & design; data collection and processing; model building/adaptation; testing/validation; deployment; operation/monitoring; and decommissioning—to organize governance, documentation, risk assessments and oversight. |
| United Nations | Principles for the Ethical Use of Artificial Intelligence in the United Nations System | Adopted | Includes procurement alongside initial research and design to use, maintenance, and termination as part of the comprehensive process. |
| Serbia | Belgrade Ministerial Declaration on Artificial Intelligence | Adopted | Frames the end-to-end sequence to guide where safety, security, governance, and accountability measures should be applied across conception through retirement. |
What it triggers
- ISO/IEC 23894:2023 describes a risk management process applied across the complete progression of an AI system from conception through retirement.source
- Switzerland's Strategy: Use of AI systems in the Federal Administration uses the structured sequence of stages to organise governance, documentation, risk assessments and oversight across an AI system's existence.source
Definitions (8)
The end-to-end stages of an AI system including data collection and curation, model training and validation, deployment, monitoring, and maintenance; used to structure obligations and risk controls across development and operational phases.
The lifecycle stages (design, train, test, deploy, monitor, retire) applied to AI systems to structure risk assessment, testing, governance and ongoing monitoring; guidance aligns these stages with OECD lifecycle concepts for operational controls.
The end-to-end sequence of stages an AI system undergoes, including conception, design, data collection, development, testing, deployment, operation, monitoring, maintenance, and eventual retirement or decommissioning. Lifecycle framing guides where safety, security, governance, and accountability measures should be applied.
The full set of phases for an AI system, including design and development, deployment, operation and monitoring, and retirement/decommissioning, encompassing activities needed to manage the system across its operational life.
A comprehensive process spanning from initial research and design to procurement, deployment, use, maintenance, and termination.
A structured sequence of stages for AI systems defined by the strategy — plan & design; data collection and processing; model building/adaptation; testing/validation; deployment; operation/monitoring; and decommissioning — used to organise governance, documentation, risk assessments and oversight across an AI system's existence.
The full range of activities and processes that are relevant to an AI system, from its conception and design to its deployment, operation, and decommissioning. This includes, but is not limited to, data collection, processing, model development, testing, validation, deployment, monitoring, and maintenance.
The complete progression of an AI system from its initial conception and design through data acquisition, model development, verification, validation, deployment, operational monitoring, incident handling, and eventual retirement. This comprehensive approach ensures that all stages of an AI system's existence are subject to structured management and oversight, particularly concerning risk.
Not to be confused with
AI lifecycle refers more concisely to stages from planning through decommissioning, while AI system lifecycle definitions detail specific phases and management activities.
AI lifecycle phases lists core stages like design, training, deployment, and monitoring, whereas AI system lifecycle encompasses the full structured progression including oversight and retirement.
Questions and answers
What stages are included in the AI system lifecycle under ISO/IEC 23894?
It covers the complete progression from initial conception and design through data acquisition, model development, verification, validation, deployment, operational monitoring, incident handling, and retirement. This ensures structured management and oversight, particularly for risks.
How does Switzerland structure the AI system lifecycle?
It defines a sequence of stages: plan & design; data collection and processing; model building/adaptation; testing/validation; deployment; operation/monitoring; and decommissioning. These organize governance, documentation, risk assessments, and oversight.
What is distinctive about the UN Principles' definition of AI system lifecycle?
It spans from initial research and design to procurement, deployment, use, maintenance, and termination. This comprehensive process guides ethical use across UN entities.
Related Terms
AI system life cycle
Phased lifecycle from design to decommissioning for AI governance....
Life cycle
The sequence of phases governing an AI system from design to decommissioning....
AI lifecycle
Stages of AI from planning through decommissioning....
AI lifecycle phases
Design, training, deployment, and monitoring stages of AI systems....
Project lifecycle
Stages from conception through operation and maintenance....