Technical

lifecycle phases

Stages of an AI system from design through retirement.

How the laws define this

Regulators consistently treat lifecycle phases as the structured, sequential, or discrete stages of an artificial intelligence system's existence to anchor proportionate risk management, governance, documentation, and continuous monitoring obligations. Assurance, assessment, and testing requirements are applied across these stages to manage risk throughout the system's life. The primary divergence lies in how granularly these stages are divided and whether end-of-life steps are addressed. The Implementation Framework for Responsible Use of Algorithms explicitly includes procurement and end-of-life data clean-up, whereas the Énoncé de principes pour une utilisation responsable de l'intelligence artificielle par les organismes publics (Québec Ministry of Cybersecurity and Digital Affairs) breaks out detailed steps like data collection and model construction or adaptation. Conversely, the Voluntary Code of Conduct on the Responsible Development and Management of Advanced Generative AI Systems outlines phases ending at monitoring, omitting the explicit retirement or decommissioning stages mandated by the National Framework for the Assurance of Artificial Intelligence in Government.

Synthesised from the 4 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.

Jurisdiction	Instrument	Status	What is distinctive
Australia	National Framework for the Assurance of Artificial Intelligence in Government	In Force	Explicitly lists six sequential stages: design, build, test, deploy, operate and retire, and emphasizes proportionate assurance activities with documentation for auditability and continuous improvement.
Quebec, Canada	Statement of Principles for the Responsible Use of Artificial Intelligence by Public Organizations	In Force	Details seven phases starting with planning and data collection through decommissioning, each with specific governance, assessment and documentation guidance.
Canada	Voluntary Code of Conduct on the Responsible Development and Management of Advanced Generative AI Systems	In Force	Uses a typical set of stages—design, development, testing, deployment and monitoring—to structure proportionate risk management, documentation, testing and oversight.
Netherlands	Implementation Framework for Responsible Use of Algorithms	Adopted	Structures five stages from design and procurement through end-of-life decommissioning and data clean-up, anchoring specific recommended measures like impact assessments and retention planning.

What it triggers

In the Australia National Framework for the Assurance of Artificial Intelligence in Government, sets out assurance activities to be applied proportionately at each lifecycle phase to manage risks and maintain documentation.source
In the Quebec Statement of Principles for the Responsible Use of Artificial Intelligence by Public Organizations, describes specific governance, assessment and documentation guidance for each lifecycle phase.source
In the Canada Voluntary Code of Conduct on the Responsible Development and Management of Advanced Generative AI Systems, uses lifecycle phases to structure proportionate risk management, documentation, testing and oversight activities.source
In the Netherlands Implementation Framework for Responsible Use of Algorithms, anchors specific recommended measures such as specifications, validation, impact assessments and continuous monitoring to each lifecycle phase.source

Definitions (4)

Lifecycle phases refer to the sequential stages of an AI system’s existence as defined in the framework: design, build, test, deploy, operate and retire. The framework requires assurance activities to be applied proportionately at each phase to manage risks and maintain documentation for auditability and continuous improvement.

National Framework for the Assurance of Artificial Intelligence in GovernmentDefinition 1 of 4

The discrete phases of an AI system's life as defined in the statement — planning, data collection, model construction or adaptation, testing and validation, deployment, operation and monitoring, and decommissioning — each of which carries specific governance, assessment and documentation obligations.

Énoncé de principes pour une utilisation responsable de l'intelligence artificielle par les organismes publics (Québec Ministry of Cybersecurity and Digital Affairs)Definition 2 of 4

A structured set of stages the framework uses to anchor obligations and responsibilities: design and procurement, development and testing, pre‑deployment assessment and approval, operational monitoring and maintenance, and end‑of‑life decommissioning and data clean‑up. Each phase carries specific recommended measures such as specifications, validation, impact assessments, continuous monitoring and retention/cleanup planning.

Implementation Framework for Responsible Use of Algorithms (Implementatiekader 'Verantwoorde inzet van algoritmen')Definition 3 of 4

Defined stages in the development and operation of an AI system—typically design, development, testing, deployment and monitoring—used by the Code to structure proportionate risk management, documentation, testing and oversight activities across the system's life.

Voluntary Code of Conduct on the Responsible Development and Management of Advanced Generative AI SystemsDefinition 4 of 4

Not to be confused with

AI lifecycle phases

The related term lists only four core stages—design, training, deployment, and monitoring—omitting phases like planning, testing, operation and decommissioning found in these definitions.

AI lifecycle

The related term broadly covers stages from planning through decommissioning, aligning closely but lacking the detailed phase lists and obligations in these definitions.

Model lifecycle

The related term focuses narrowly on model-specific phases from design and data curation to decommissioning, excluding broader system stages like deployment and operation.

Questions and answers

What stages make up lifecycle phases in the Australian AI assurance framework?

It defines six sequential stages: design, build, test, deploy, operate and retire. Assurance activities are applied proportionately at each to manage risks.

How does Quebec's AI principles statement handle lifecycle phases?

It outlines seven phases from planning and data collection to decommissioning. Each phase includes specific governance, assessment and documentation guidance.

What measures does the Netherlands framework recommend for AI lifecycle phases?

It structures five stages including design, development, pre-deployment assessment and end-of-life data clean-up. Each phase includes recommended measures like validation, impact assessments and monitoring.