AIGO — AI Governance Operating Framework
AI Retirement Procedure
Version: 0.1Status: Draft
Working Name: AIGO
Full Name: AI Governance Operating Framework
Document Identifier: AIGO-PROC-012
1. Purpose
This procedure defines the process for planning, approving, executing, documenting, and verifying the retirement of AI systems. The procedure ensures that AI systems are removed from operational use in a controlled manner and that associated risks, data, dependencies, access, controls, records, and obligations are appropriately addressed.2. Scope
This procedure applies to the retirement of AI systems within the organization’s AIGO governance scope. It may apply to:- AI models;
- AI applications;
- generative AI systems;
- AI agents;
- AI-enabled business processes;
- AI services;
- third-party AI services;
- AI infrastructure;
- datasets and data pipelines; and
- supporting governance mechanisms.
3. Objectives
The objectives of AI retirement are to:- ensure retirement is authorized;
- prevent unintended continued operation;
- manage transition risks;
- protect data;
- revoke unnecessary access;
- manage dependencies;
- preserve required records;
- address residual risks;
- meet applicable obligations; and
- formally confirm retirement.
4. Retirement Principles
AI retirement should be:- planned;
- authorized;
- controlled;
- documented;
- traceable;
- risk-based;
- proportionate;
- reversible where appropriate; and
- verified.
5. Retirement Triggers
An AI system may be considered for retirement when:- its intended purpose is no longer required;
- a replacement system is available;
- the system is obsolete;
- risk has become unacceptable;
- performance is no longer adequate;
- the supporting technology is no longer available;
- a supplier service is being discontinued;
- regulatory requirements change;
- the system is no longer economically viable; or
- governance requirements cannot reasonably be maintained.
6. Retirement Request
A retirement request should identify:- AI system;
- system owner;
- reason for retirement;
- proposed retirement date;
- dependencies;
- affected stakeholders;
- replacement arrangements;
- risk considerations;
- data considerations;
- contractual obligations; and
- required approvals.
7. Retirement Assessment
Before retirement, the organization should assess:- operational impact;
- business continuity;
- security;
- privacy;
- legal and regulatory requirements;
- contractual obligations;
- data retention;
- records;
- dependencies;
- user impact;
- risk; and
- replacement arrangements.
8. Retirement Classification
The retirement process should be proportionate to the significance of the AI system. Factors may include:- system criticality;
- risk classification;
- number of users;
- affected stakeholders;
- regulatory significance;
- autonomy;
- data sensitivity;
- third-party dependencies; and
- business impact.
9. Retirement Approval
Retirement should receive approval from an appropriately authorized authority. Approval may involve:- AI system owner;
- business owner;
- AI governance;
- risk management;
- security;
- privacy;
- legal;
- compliance; and
- executive management.
10. Retirement Plan
A retirement plan should be established for material systems. The plan should identify:- activities;
- responsibilities;
- dependencies;
- timeline;
- communication;
- transition;
- data handling;
- access removal;
- system shutdown;
- validation; and
- closure.
11. Stakeholder Identification
Relevant stakeholders should be identified before retirement. Stakeholders may include:- users;
- customers;
- employees;
- system owners;
- suppliers;
- affected business functions;
- governance functions; and
- regulators where applicable.
12. Communication
Material retirement should be communicated to affected stakeholders. Communication may include:- retirement date;
- reason;
- expected impact;
- replacement arrangements;
- transition requirements;
- changes in responsibilities; and
- support arrangements.
13. Replacement Systems
Where an AI system is being replaced, the organization should assess the replacement system separately according to applicable AIGO requirements. Retirement approval should not automatically constitute approval of the replacement system.14. Transition Planning
Transition should be planned where operational continuity is required. Transition arrangements may include:- replacement AI system;
- alternative technology;
- manual process;
- temporary service;
- phased migration; or
- controlled service reduction.
15. Dependency Assessment
Dependencies should be identified before retirement. Dependencies may include:- applications;
- APIs;
- datasets;
- models;
- infrastructure;
- workflows;
- suppliers;
- monitoring;
- controls; and
- human processes.
16. Dependency Removal
Dependencies that are no longer required should be safely removed or disabled. The organization should verify that removal does not unintentionally affect other systems.17. Data Assessment
Data associated with the AI system should be assessed before retirement. The assessment should consider:- data ownership;
- retention requirements;
- legal obligations;
- contractual requirements;
- business records;
- personal data;
- sensitive data; and
- deletion requirements.
18. Data Retention
Required records and data should be retained according to applicable requirements. Retention should distinguish between:- records that must be retained;
- data that may be retained;
- data that should be deleted; and
- data that should be securely archived.
19. Data Deletion
Where deletion is required, it should be performed using appropriate methods. Deletion should consider:- production data;
- backups;
- replicas;
- caches;
- logs;
- exported data;
- third-party copies; and
- derived data.
20. Personal Data
Where personal data is involved, retirement should include an appropriate privacy assessment. The organization should determine:- what personal data remains;
- retention requirements;
- deletion requirements;
- access requirements;
- archival requirements; and
- third-party obligations.
21. Security Assessment
Before retirement, security requirements should be reviewed. The assessment may include:- credentials;
- accounts;
- service identities;
- keys;
- certificates;
- network access;
- integrations;
- privileged access; and
- security monitoring.
22. Access Revocation
Access that is no longer required should be revoked. This may include:- user accounts;
- service accounts;
- API credentials;
- tokens;
- keys;
- privileged access; and
- external integrations.
23. Infrastructure Retirement
Supporting infrastructure should be identified and appropriately decommissioned. Infrastructure may include:- servers;
- containers;
- databases;
- storage;
- networks;
- APIs;
- cloud resources; and
- monitoring infrastructure.
24. Model Retirement
Where a model is retired, the organization should identify:- model versions;
- deployment locations;
- stored artifacts;
- dependent systems;
- evaluation records;
- monitoring records; and
- retention requirements.
25. AI Agent Retirement
Where an AI agent is retired, the organization should verify that:- autonomous execution is disabled;
- tool access is removed;
- credentials are revoked;
- scheduled actions are disabled;
- integrations are removed; and
- outstanding tasks are resolved.
26. Third-Party Service Retirement
Where a third-party AI service is retired, the organization should address:- service termination;
- contractual requirements;
- data return;
- data deletion;
- account closure;
- credential revocation;
- integration removal; and
- evidence of provider actions where required.
27. Monitoring Changes
Monitoring requirements should be updated when the system is retired. The organization should ensure that:- obsolete alerts are removed;
- monitoring dependencies are removed;
- required historical records are preserved; and
- monitoring coverage remains appropriate for replacement systems.
28. Control Changes
Controls associated exclusively with the retired system should be reviewed. Controls should be:- retired;
- transferred;
- modified; or
- retained
29. Risk Review
The risk assessment should be reviewed before retirement. The review should consider:- retirement risks;
- transition risks;
- residual risks;
- data risks;
- security risks;
- privacy risks;
- continuity risks; and
- replacement risks.
30. Risk Closure
Risks that exist only because of the retired system may be closed when sufficient evidence confirms that the underlying exposure no longer exists. Risks transferred to another system or process should remain open and be reassigned appropriately.31. Incident Review
Open incidents involving the AI system should be reviewed before retirement. The organization should determine whether incidents:- must be resolved before retirement;
- can be transferred;
- require continued investigation; or
- require retention of evidence.
32. Legal and Regulatory Review
Retirement should consider applicable:- laws;
- regulations;
- regulatory reporting;
- contractual obligations;
- litigation holds;
- investigations;
- records requirements; and
- other legal obligations.
33. Records Management
Governance records should be retained according to applicable requirements. Records may include:- system profile;
- risk assessments;
- approvals;
- change records;
- monitoring records;
- incidents;
- assurance reports;
- audit evidence;
- contracts; and
- retirement evidence.
34. Retirement Testing
Before final shutdown, appropriate testing should confirm that:- dependencies have been identified;
- transition is functional;
- required data has been handled;
- access has been reviewed;
- integrations are understood; and
- shutdown will not create unacceptable impact.
35. Shutdown
The system should be shut down according to the approved retirement plan. Shutdown may involve:- disabling access;
- disabling deployment;
- stopping services;
- removing integrations;
- disabling scheduled tasks;
- decommissioning infrastructure; and
- removing unnecessary credentials.
36. Retirement Validation
After shutdown, the organization should validate that the AI system is no longer operating in unauthorized or unintended ways. Validation may include:- access testing;
- service checks;
- integration checks;
- credential checks;
- monitoring review;
- infrastructure review; and
- user confirmation.
37. Residual Operation
The organization should assess whether any components continue to operate after retirement. Examples may include:- background services;
- scheduled jobs;
- cached models;
- APIs;
- replicas;
- automated workflows; and
- third-party integrations.
38. Retirement Evidence
Evidence should be maintained to demonstrate retirement. Evidence may include:- approval;
- retirement plan;
- communication;
- access revocation;
- data handling;
- shutdown records;
- infrastructure decommissioning;
- validation;
- supplier confirmation; and
- closure approval.
39. Retirement Closure
Retirement should be formally closed when:- approved shutdown is complete;
- required data actions are complete;
- access has been addressed;
- dependencies have been resolved;
- monitoring has been updated;
- risks have been addressed;
- records have been preserved; and
- validation is complete.
40. Retirement Register
The organization should maintain a record of retired AI systems. The register may contain:- system identifier;
- system name;
- owner;
- retirement reason;
- retirement date;
- replacement system;
- residual risks;
- records location;
- approval; and
- closure status.
41. Retirement Reporting
The AI governance function may report retirement activity to appropriate governance bodies. Reporting may include:- systems retired;
- systems scheduled for retirement;
- delayed retirements;
- retirement risks;
- replacement systems;
- outstanding dependencies; and
- unresolved residual risks.
42. Retirement Metrics
Organizations may establish retirement metrics. Examples include:- retirement completion rate;
- overdue retirements;
- unresolved dependencies;
- incomplete access removal;
- outstanding data actions;
- residual risks; and
- retirement-related incidents.
43. Retirement and Change Management
Retirement should be managed as a controlled lifecycle change. The retirement activity should comply with applicable change management requirements.44. Retirement and Risk Management
Retirement should update relevant risk records. Risks transferred to replacement systems or alternative processes should be reassessed and assigned appropriately.45. Retirement and Incident Management
Retirement should account for unresolved incidents. Where evidence must be retained, it should remain accessible after the system has been decommissioned.46. Retirement and Assurance
Material retirements may be subject to assurance or independent review where appropriate. Assurance may verify:- authorization;
- data handling;
- access removal;
- dependency removal;
- risk closure;
- evidence; and
- final shutdown.
47. Responsibilities
AI System Owner- initiate retirement;
- develop the retirement plan;
- coordinate stakeholders;
- ensure required actions are completed; and
- confirm retirement.
- assess business impact;
- approve transition arrangements;
- coordinate affected users; and
- confirm operational requirements.
- oversee governance requirements;
- review material retirements;
- ensure traceability;
- monitor retirement status; and
- confirm governance closure.
- execute shutdown;
- remove infrastructure and integrations;
- revoke technical access; and
- provide technical evidence.
- provide specialist review where required;
- identify obligations;
- verify required actions; and
- support closure.
48. Retirement Workflow
The standard workflow should be:- Identify retirement trigger.
- Initiate retirement request.
- Assess system and business impact.
- Identify stakeholders and dependencies.
- Assess risks.
- Assess data and records.
- Assess security and privacy.
- Assess legal and regulatory requirements.
- Identify replacement or transition arrangements.
- Develop retirement plan.
- Obtain required approvals.
- Communicate retirement.
- Implement transition.
- Disable or decommission the AI system.
- Remove unnecessary access and dependencies.
- Complete required data actions.
- Validate shutdown.
- Update governance records.
- Review residual risks and incidents.
- Obtain closure approval.
- Record lessons learned.
- Close retirement.
49. Continuous Improvement
The retirement process should be improved based on:- retirement experience;
- failed transitions;
- incidents;
- assurance findings;
- audit findings;
- recurring dependencies;
- stakeholder feedback;
- technology changes; and
- operational experience.
50. Procedure Review
This procedure should be reviewed periodically and when material changes occur. Review triggers may include:- significant retirement incidents;
- changes to AIGO requirements;
- regulatory developments;
- changes in technology;
- assurance findings;
- organizational changes; and
- implementation experience.
51. Procedure Status
Document: AIGO AI Retirement Procedure Version: 0.1 Status: Draft Working Name: AIGO Full Name: AI Governance Operating Framework Document Identifier:AIGO-PROC-012
Document Type: Operational Procedure
This procedure establishes the operational process for safely and formally retiring AI systems and closing or transferring their associated governance obligations.