Risk · R-102 · Governance & Accountability
Uncontrolled AI retirement
A system is withdrawn without managing dependent processes, retained data, model artefacts and continuing obligations.
- Severity
- Medium
- Audience
- Practitioner
- Frameworks
- Principle 13 · Annex A.6 · Principle 3
- Responsible AI domain
- Lifecycle stage
- Reviewed
- 2026-11-10
What it is
Retirement is a stage of the lifecycle with its own risks. Processes that relied on the system need a replacement. Copies of training data, model files, embeddings and logs persist unless removed. Credentials remain live. Duties continue after the system has gone, such as explaining past decisions and keeping records for a set period. The risk is that a system is simply switched off, or left running unowned after its replacement arrives.
How it happens
- A supplier withdraws a product at short notice.
- The replacement goes live and nobody decommissions the old system.
- Data and artefacts are left in storage with no owner.
- Records needed to explain past decisions are deleted with everything else.
Why it matters
Abandoned data and credentials are a security and data protection exposure. Dependent processes fail or carry on using stale outputs. When a past decision is later challenged, the means of explaining it no longer exist.
Warning signs
- The inventory lists systems nobody uses.
- Storage contains model files and datasets with no owner.
- Credentials for retired systems remain active.
- No procedure describes how a system is withdrawn.
Mitigating controls
Primary controls
- AI decommissioning procedure Directive
- Plans retirement to cover dependants, data, artefacts and continuing duties. Retirement of an AI system follows a plan that covers dependent processes, disposal of data and model artefacts, retention of records and any continuing obligations.
Supporting controls
- AI inventory and registration Detective
- Shows what depends on the system and updates its status. All AI systems and models, whether built, bought or embedded in other products, are recorded with owner, purpose, tier, data, supplier and status, and the record is reconciled periodically.
- Prompt and log retention limits Preventive
- Governs what happens to retained data. Prompts, outputs and conversation histories are kept for defined periods and then deleted, by the organisation and by its providers, with deletion verified.
Framework and clause references
| Framework | Clause | Title |
|---|---|---|
| Code of Practice for the Cyber Security of AI | Principle 13 | Ensure proper data and model disposal Disposal of data and models. |
| ISO/IEC 42001:2023 | Annex A.6 | AI system life cycle Retirement within the lifecycle. |
| SS1/23: Model Risk Management Principles for Banks | Principle 3 | Model development, implementation and use Decommissioning of models. |
Sources
- SS1/23: Model Risk Management Principles for Banks Bank of England PRA · official text
- ISO/IEC 42001:2023 – AI management systems ISO/IEC · framework
How to cite
Axiom Verity (2026). Uncontrolled AI retirement. Reference Library, R-102. https://axiom-verity.com/risks/uncontrolled-ai-retirement. Reviewed 2026-11-10.
Found an error or a better source? Report a correction.