← 3D visual tours

AIGP study · Domains III and IV

The life cycle, and where governance sits

An AI system is not a single event. It moves through stages, and the governance questions change at each one. This page walks through a seven-stage view, with GOVERN running through all of them.

Sources: BoK v2.1 Domain III and IV competency titles; NIST AI RMF 1.0.

The seven stages are a study model. Real projects overlap, loop back and split. The mapping to NIST functions is a common study mapping, not an official one.

Stage 1 · Plan and design

Decide what you are building, and why

Before any model work, write down the purpose, the people the system will touch and the setting it will run in. Early context-setting is the part of the NIST framework labelled MAP.

Related BoK competency: III.A (governing design and build). Source: NIST AI RMF 1.0, MAP function.

Stage 2 · Data

Collect and prepare the data

Training and test data bring their own questions about quality, bias and privacy. Personal data needs a lawful basis under GDPR Article 6, and sensitive categories need a condition under Article 9.

Related BoK competency: III.B (governing data collection and use). Sources: GDPR Art. 6 and 9.

Stage 3 · Build and train

Make the design choices visible

Record architecture choices, training runs and the reasons behind them as you go. For high-risk systems, the AI Act expects technical documentation, and it is far easier to write that as you build than to reconstruct it later.

Related BoK competency: III.A. Source: AI Act Art. 11.

Stage 4 · Test and validate

Check before anyone relies on it

Test accuracy, robustness and fairness against measures you defined in advance, and keep the results. For high-risk systems, risk management under Article 9 covers the whole cycle, not just this stage.

Related BoK competency: IV.B (assessing the AI system). Sources: AI Act Art. 9; NIST AI RMF MEASURE function.

Stage 5 · Deploy

Put it into the real process

Decide who uses the system, for what, and with what human oversight. Watch for automation bias, the tendency to accept a system's output without enough checking. The Act names this risk for human oversight duties (Article 14(4)(b)).

Related BoK competency: IV.C (governing deployment and use). Source: AI Act Art. 14(4)(b).

Stage 6 · Operate and monitor

Once live, the system keeps changing

Data drifts, users adapt and incidents happen. Monitor performance, keep logs and report serious incidents on the timelines the Act sets. For providers of high-risk systems, post-market monitoring is required.

Related BoK competency: III.C (release, monitoring and maintenance). Sources: AI Act Art. 72 and Art. 73.

Stage 7 · Retire

Plan the end, too

Retiring a system means deciding what happens to its data, its logs, its models and the people who relied on it. It is easy to forget, so it has its own stage here. Your lecture notes tagged D4 cover decommissioning, so read them alongside this chapter.

Related BoK competency: III.C. Source: lecture notes D4.

Across all seven · GOVERN

GOVERN is not a stage

Policies, roles, accountability, risk tolerance and training run through every stage. In the NIST framework, GOVERN is the cross-cutting function. The BoK sets the same expectations at organisation level in competencies I.B and I.C.

Sources: NIST AI RMF 1.0, GOVERN function; BoK I.B and I.C.