Portfolio & Use Cases ยท POR-09
Retiring uses of AI
Uses of AI that no longer earn their place are withdrawn through a planned process that covers users, dependent processes and disposal or retention of models and data.
Why it matters
AI that stays in use after it stops delivering costs money, carries risk and may give worse results as conditions change. Unplanned withdrawal leaves users stranded and leaves models and data behind with nobody accountable for them.
Levels
| 0 Absent | Confirmed absent: the evidence shows nothing is in place. Not known is recorded as insufficient evidence, not 0. |
|---|---|
| 1 Initial | Uses are left running or switched off without a plan; models and data are left in place. |
| 2 Developing | Some uses are retired deliberately, but steps for users, data and models are worked out each time. |
| 3 Defined | A documented retirement process, triggered by periodic review, withdraws uses of AI with care for users, people, data, models and hardware, and updates the inventory. |
| 4 Managed | Live uses are routinely reviewed against retirement triggers, retirements are checked for completion, and leftover models or data are found and dealt with. |
| 5 Optimising | Retirement is planned from the design stage, and the process is refined from lessons of past withdrawals. |
Level 3 criteria
Level 3 is met when every applicable criterion is met. Criteria marked with an overlay apply only to organisations with that characteristic (overlays).
- POR-09.1 A documented retirement process is triggered by periodic review of live uses of AI.
- POR-09.2 The retirement process covers user transition, dependent processes, records, and disposal or retention of models and data.
- POR-09.3 The retirement process covers reuse, recycling or disposal of dedicated hardware and infrastructure.
- POR-09.4 Each retirement updates the AI inventory (GOV-06).
- POR-09.5 The effect of a withdrawal on affected people and services is assessed, and groups at greater risk are identified.
- POR-09.6 Users and affected people are told about a withdrawal in advance.
Assessment questions
- How do you decide that a use of AI should be retired?
- When a use was last retired, what happened to its users, models and data?
- Are there uses still running that nobody would miss?
- Withdrawal impact assessment and notice to users and affected people for each retired use
- The retirement process and its triggers
- Records of a completed retirement, including data and model disposal
- AI inventory entries marked as retired
How to reach Level 3
- Set a periodic review of live uses that asks whether each still earns its place and triggers retirement if not.
- Write a retirement process covering user transition, dependent processes, records, disposal or retention of models and data, and reuse, recycling or disposal of dedicated hardware and infrastructure.
- Assess the effect of withdrawal on affected people and services, identify groups at greater risk, and tell users and affected people in advance.
- Update the AI inventory (GOV-06) when a use is retired.
Common pitfalls
- Switching off a use without checking which processes depend on it.
- Keeping models and data indefinitely after retirement with no retention decision (see DAT-06).
Starter artefacts
- AI retirement checklist
- Withdrawal impact and communication plan
Cross-references
| AI Cyber Security Code of Practice | 13.1 partial; 13.2 direct |
|---|---|
| Algorithmic Transparency Recording Standard | Scope timing and updates partial |
| GDS AI Playbook | P5 partial |
| JSP 936 Part 1 | para 15 partial |
| NIST AI RMF 1.0 | GOVERN 1.7 direct; MANAGE 4.1 partial |
| UK Data and AI Ethics Framework | Privacy 11 direct; Societal impact 6 direct; SAT 4.10 direct; SAT 4.11 direct; SAT 4.13 direct |
| UNESCO AI Maturity Framework | 4.1 partial |
direct: this control at Level 3 meets the item. partial: it covers part of it.