Skip to main content

AIGO — AI Governance Operating Framework

AI Governance Lifecycle

Version: 0.1
Status: Draft
Working Name: AIGO
Full Name: AI Governance Operating Framework

1. Purpose

The AIGO AI Governance Lifecycle defines the governance activities required to manage AI systems throughout their organizational lifecycle. The lifecycle provides a structured approach for governing AI systems from initial identification and conception through assessment, design, development, testing, approval, deployment, operation, monitoring, change, and retirement. The lifecycle is intended to provide continuity of governance rather than treating AI governance as a single approval event. Organizations should apply lifecycle activities proportionately according to:
  • the characteristics of the AI system;
  • intended use;
  • risk;
  • impact;
  • level of autonomy;
  • organizational context;
  • applicable requirements; and
  • AI System Profile.

2. Lifecycle Principles

The AIGO AI Governance Lifecycle is based on the following principles:
  • governance should begin before material AI deployment decisions are made;
  • governance should continue throughout the AI system lifecycle;
  • governance activities should be proportionate to risk;
  • significant decisions should be documented;
  • responsibilities should remain clearly assigned;
  • changes should trigger appropriate reassessment;
  • monitoring should continue after deployment;
  • evidence should be maintained throughout the lifecycle;
  • significant incidents should trigger appropriate governance response; and
  • retirement should be treated as a governed lifecycle activity.

3. Lifecycle Model

AIGO defines the following primary lifecycle stages:
  1. Identification and Initiation
  2. Initial Assessment
  3. Concept and Purpose Definition
  4. Governance Classification
  5. Risk Assessment
  6. Design and Planning
  7. Development and Acquisition
  8. Validation and Testing
  9. Governance Review and Approval
  10. Deployment and Transition
  11. Operation and Monitoring
  12. Change and Reassessment
  13. Incident and Issue Management
  14. Periodic Review
  15. Retirement and Decommissioning
  16. Post-Retirement Review
The stages may be applied sequentially, iteratively, or with appropriate tailoring according to the nature of the AI system.

4. Lifecycle Stage Governance

Each lifecycle stage should define, as appropriate:
  • purpose;
  • entry criteria;
  • required activities;
  • responsible roles;
  • required decisions;
  • risk considerations;
  • required controls;
  • required evidence;
  • exit criteria; and
  • escalation requirements.
Organizations may combine lifecycle stages where doing so does not reduce governance effectiveness.

5. Stage 1 — Identification and Initiation

5.1 Purpose

The Identification and Initiation stage establishes whether an AI initiative, system, capability, or service should enter the organization’s AI governance lifecycle.

5.2 Initiation Triggers

An AI governance lifecycle may be initiated when an organization:
  • proposes a new AI system;
  • acquires an external AI service;
  • introduces AI into an existing application;
  • introduces a new AI model;
  • introduces an AI agent;
  • introduces an agentic workflow;
  • materially changes an existing AI system;
  • identifies previously ungoverned AI use;
  • discovers unauthorized AI use; or
  • determines that an existing system requires formal governance.

5.3 Initial Information

The organization should collect sufficient initial information to determine the appropriate governance path. Information may include:
  • proposed system name;
  • business purpose;
  • intended users;
  • system owner;
  • business owner;
  • technology provider;
  • proposed model or service;
  • intended data;
  • expected outputs;
  • intended actions;
  • deployment environment;
  • level of autonomy;
  • affected stakeholders; and
  • anticipated use cases.

5.4 Entry Criteria

The stage may begin when there is sufficient information to identify the proposed AI activity and its organizational owner.

5.5 Exit Criteria

The stage should result in:
  • an identified AI initiative;
  • an identified accountable owner;
  • an initial description of intended use;
  • an initial governance classification; and
  • a determination of whether further assessment is required.

6. Stage 2 — Initial Assessment

6.1 Purpose

The Initial Assessment determines whether the proposed AI system requires additional governance activities and establishes an initial understanding of its risk and impact.

6.2 Assessment Areas

The assessment may consider:
  • intended purpose;
  • affected stakeholders;
  • type of AI capability;
  • level of autonomy;
  • data involved;
  • potential impact;
  • security considerations;
  • privacy considerations;
  • legal and regulatory considerations;
  • third-party dependencies;
  • operational dependencies; and
  • potential misuse.

6.3 Initial Risk Screening

Organizations should perform an initial risk screening sufficient to determine the appropriate governance pathway. The screening should consider whether the system may create:
  • significant business impact;
  • material financial impact;
  • significant security risk;
  • privacy risk;
  • legal or regulatory risk;
  • safety risk;
  • reputational risk;
  • discrimination or fairness concerns;
  • human rights concerns;
  • operational risk; or
  • other material impacts.

6.4 Assessment Outcome

The outcome may include:
  • proceed with standard governance;
  • proceed with enhanced governance;
  • require additional specialist assessment;
  • require management approval;
  • defer the initiative; or
  • prohibit the proposed use.

7. Stage 3 — Concept and Purpose Definition

7.1 Purpose

The Concept and Purpose Definition stage establishes the intended purpose, scope, objectives, and boundaries of the AI system.

7.2 Intended Purpose

The intended purpose should be documented clearly enough to determine:
  • what the system is intended to do;
  • who is expected to use it;
  • what decisions or activities it supports;
  • what actions it may perform;
  • what it is not intended to do; and
  • what conditions apply to its use.

7.3 Intended Use

The organization should distinguish between:
  • intended use;
  • reasonably foreseeable use;
  • permitted use;
  • prohibited use; and
  • unsupported use.

7.4 Purpose Boundaries

Purpose boundaries should identify relevant limitations on:
  • users;
  • data;
  • outputs;
  • decisions;
  • actions;
  • environments;
  • integrations; and
  • system autonomy.

8. Stage 4 — Governance Classification

8.1 Purpose

The Governance Classification stage determines the level and type of governance required for the AI system.

8.2 Classification Factors

Classification may consider:
  • system purpose;
  • business criticality;
  • potential impact;
  • affected individuals;
  • level of automation;
  • level of autonomy;
  • data sensitivity;
  • security exposure;
  • regulatory exposure;
  • model characteristics;
  • third-party dependency; and
  • operational complexity.

8.3 Governance Categories

Organizations may establish governance categories such as:
  • standard;
  • elevated;
  • high;
  • critical; or
  • another risk-based classification model.
The exact categories may be defined by the organization’s governance policy.

8.4 Classification Review

Classification should be reviewed when:
  • intended use changes;
  • system capabilities change;
  • autonomy increases;
  • new data is introduced;
  • the operating environment changes;
  • material incidents occur; or
  • applicable requirements change.

9. Stage 5 — Risk Assessment

9.1 Purpose

The Risk Assessment stage identifies, evaluates, and documents risks associated with the AI system.

9.2 Risk Identification

Risk identification should consider, where relevant:
  • organizational risks;
  • operational risks;
  • technical risks;
  • cybersecurity risks;
  • privacy risks;
  • data risks;
  • legal and regulatory risks;
  • ethical risks;
  • safety risks;
  • financial risks;
  • reputational risks;
  • third-party risks;
  • human factors; and
  • societal or stakeholder impacts.

9.3 Risk Analysis

Risks should be analyzed using an appropriate organizational methodology. The analysis may consider:
  • likelihood;
  • impact;
  • exposure;
  • affected stakeholders;
  • existing controls;
  • residual risk; and
  • uncertainty.

9.4 Risk Treatment

Risk treatment may include:
  • mitigation;
  • avoidance;
  • transfer;
  • acceptance;
  • additional controls;
  • restricted use; or
  • discontinuation.

9.5 Risk Acceptance

Residual risk should be accepted only by an appropriately authorized role. Risk acceptance should be documented where material.

10. Stage 6 — Design and Planning

10.1 Purpose

The Design and Planning stage translates governance, business, risk, security, privacy, and operational requirements into system design and implementation requirements.

10.2 Design Considerations

Design activities may consider:
  • system architecture;
  • model selection;
  • data architecture;
  • security architecture;
  • privacy requirements;
  • human oversight;
  • access control;
  • monitoring;
  • logging;
  • resilience;
  • fail-safe behavior;
  • intervention mechanisms;
  • integration boundaries; and
  • third-party dependencies.

10.3 Governance by Design

Governance requirements should be incorporated into system design where practical rather than added only after implementation.

10.4 Design Evidence

Design evidence may include:
  • architecture documentation;
  • data-flow documentation;
  • risk assessments;
  • security assessments;
  • privacy assessments;
  • requirements specifications;
  • model documentation;
  • control specifications; and
  • testing plans.

11. Stage 7 — Development and Acquisition

11.1 Purpose

The Development and Acquisition stage covers the development, configuration, integration, procurement, or acquisition of the AI system.

11.2 Internal Development

Internal development should follow applicable organizational engineering and governance processes. Activities may include:
  • requirements implementation;
  • model development;
  • application development;
  • data preparation;
  • integration;
  • testing;
  • security implementation;
  • monitoring implementation; and
  • documentation.

11.3 Third-Party Acquisition

Where an AI system or service is acquired from a third party, organizations should conduct appropriate due diligence. Considerations may include:
  • provider capabilities;
  • security;
  • privacy;
  • reliability;
  • contractual terms;
  • data handling;
  • model limitations;
  • transparency;
  • service dependencies;
  • change management; and
  • exit arrangements.

11.4 Development Evidence

Evidence should be maintained according to the system’s governance requirements. Examples include:
  • source and configuration records;
  • model information;
  • test results;
  • technical documentation;
  • supplier documentation;
  • security evidence;
  • privacy evidence; and
  • approval records.

12. Stage 8 — Validation and Testing

12.1 Purpose

The Validation and Testing stage determines whether the AI system performs sufficiently for its intended purpose and whether applicable governance requirements have been satisfied.

12.2 Testing Scope

Testing should be proportionate to risk and may include:
  • functional testing;
  • performance testing;
  • accuracy testing;
  • robustness testing;
  • security testing;
  • privacy testing;
  • bias and fairness assessment;
  • misuse testing;
  • adversarial testing;
  • human factors testing;
  • safety testing; and
  • operational testing.

12.3 Test Evidence

Test results should be documented sufficiently to demonstrate:
  • what was tested;
  • how it was tested;
  • expected results;
  • actual results;
  • identified deficiencies;
  • remediation activities; and
  • approval or disposition.

12.4 Deficiencies

Material deficiencies should be:
  • documented;
  • risk assessed;
  • remediated where required;
  • accepted by an authorized role; or
  • used as a basis for restricting or preventing deployment.


13. Stage 9 — Governance Review and Approval

13.1 Purpose

The Governance Review and Approval stage determines whether the AI system has satisfied the governance requirements necessary for deployment or the next applicable lifecycle stage.

13.2 Review Scope

The review should consider, as applicable:
  • intended purpose;
  • governance classification;
  • risk assessment;
  • risk treatment;
  • applicable controls;
  • security assessment;
  • privacy assessment;
  • legal and regulatory assessment;
  • technical validation;
  • human oversight;
  • operational readiness;
  • third-party requirements;
  • required evidence; and
  • outstanding issues.

13.3 Approval Decision

The approval decision may result in:
  • approval;
  • conditional approval;
  • approval with restrictions;
  • deferral pending remediation;
  • rejection; or
  • escalation to a higher authority.

13.4 Approval Authority

Approval should be provided by an appropriately authorized role according to:
  • governance classification;
  • risk level;
  • organizational policy;
  • system impact; and
  • applicable requirements.

13.5 Conditions of Approval

Where approval is conditional, the conditions should be documented. Conditions may include:
  • restricted users;
  • restricted use cases;
  • additional monitoring;
  • additional human oversight;
  • additional security controls;
  • remediation deadlines;
  • additional assurance; or
  • limitations on system autonomy.

14. Stage 10 — Deployment and Transition

14.1 Purpose

The Deployment and Transition stage moves the AI system from development or testing into its approved operational environment.

14.2 Deployment Readiness

Before deployment, organizations should confirm, as appropriate:
  • required approvals are complete;
  • identified material risks are addressed;
  • required controls are implemented;
  • monitoring is operational;
  • security controls are active;
  • access controls are configured;
  • human oversight is established;
  • incident procedures are available;
  • users are appropriately trained;
  • required documentation is available; and
  • operational ownership has been established.

14.3 Controlled Deployment

Deployment should be performed according to an approved deployment process. For higher-risk systems, organizations may use controlled deployment approaches such as:
  • pilot deployment;
  • phased rollout;
  • limited user groups;
  • restricted environments;
  • additional monitoring; or
  • progressive expansion.

14.4 Transition to Operations

The transition should establish clear responsibility for:
  • system operation;
  • monitoring;
  • incident response;
  • maintenance;
  • access management;
  • change management; and
  • lifecycle documentation.

15. Stage 11 — Operation and Monitoring

15.1 Purpose

The Operation and Monitoring stage ensures that the AI system continues to operate within approved conditions throughout its operational lifecycle.

15.2 Operational Monitoring

Monitoring may include:
  • system availability;
  • performance;
  • accuracy;
  • output quality;
  • security events;
  • privacy events;
  • unusual behavior;
  • model or data drift;
  • user activity;
  • access activity;
  • control effectiveness; and
  • significant changes in risk.

15.3 Monitoring Frequency

Monitoring frequency should be proportionate to:
  • system risk;
  • system criticality;
  • level of autonomy;
  • operational impact;
  • rate of change; and
  • applicable requirements.

15.4 Monitoring Evidence

Monitoring activities should produce appropriate evidence. Evidence may include:
  • monitoring records;
  • alerts;
  • incident records;
  • performance reports;
  • review records;
  • control results;
  • audit logs; and
  • risk indicators.

15.5 Operational Exceptions

Material deviations from approved operation should be:
  • investigated;
  • documented;
  • risk assessed;
  • remediated; and
  • escalated where appropriate.

16. Stage 12 — Change and Reassessment

16.1 Purpose

The Change and Reassessment stage governs material changes to an AI system and determines whether previous governance decisions remain valid.

16.2 Changes Requiring Assessment

Changes may include:
  • model replacement;
  • model version changes;
  • significant model updates;
  • changes to training data;
  • changes to retrieval sources;
  • changes to prompts or system instructions;
  • changes to tools;
  • changes to integrations;
  • changes to permissions;
  • changes to intended use;
  • changes to users;
  • changes to deployment environment;
  • increases in autonomy; or
  • changes to third-party providers.

16.3 Change Classification

Changes should be classified according to their potential impact. Classification may determine whether the change requires:
  • routine approval;
  • technical review;
  • risk reassessment;
  • security review;
  • privacy review;
  • legal review;
  • governance committee review; or
  • reapproval of the AI system.

16.4 Reassessment

A material change should trigger reassessment of relevant:
  • risks;
  • controls;
  • governance classification;
  • intended purpose;
  • system profile;
  • evidence requirements; and
  • approval conditions.

17. Stage 13 — Incident and Issue Management

17.1 Purpose

The Incident and Issue Management stage provides governance for unexpected, harmful, unauthorized, or materially non-compliant AI system behavior.

17.2 AI Governance Incidents

AI governance incidents may include:
  • unauthorized AI use;
  • material security events;
  • privacy incidents;
  • harmful outputs;
  • significant model failures;
  • unexpected autonomous actions;
  • control failures;
  • significant bias or discrimination concerns;
  • material data issues;
  • regulatory concerns; and
  • significant third-party failures.

17.3 Incident Response

Organizations should establish appropriate procedures for:
  • detection;
  • reporting;
  • triage;
  • containment;
  • investigation;
  • remediation;
  • escalation;
  • communication;
  • evidence preservation; and
  • lessons learned.

17.4 Temporary Restrictions

Where appropriate, organizations may:
  • restrict system functionality;
  • restrict users;
  • disable integrations;
  • suspend autonomous actions;
  • place the system into a controlled state; or
  • suspend the system entirely.

17.5 Post-Incident Review

Material incidents should be reviewed to determine whether changes are required to:
  • risks;
  • controls;
  • system design;
  • monitoring;
  • training;
  • procedures;
  • governance roles; or
  • lifecycle requirements.

18. Stage 14 — Periodic Review

18.1 Purpose

The Periodic Review stage evaluates whether the AI system remains appropriate, effective, controlled, and aligned with its approved purpose.

18.2 Review Areas

Periodic review may consider:
  • continued business purpose;
  • system performance;
  • risk profile;
  • control effectiveness;
  • security;
  • privacy;
  • regulatory requirements;
  • incidents;
  • user feedback;
  • stakeholder impact;
  • third-party changes;
  • model changes; and
  • governance maturity.

18.3 Review Frequency

Review frequency should be proportionate to the AI system’s risk and characteristics. Higher-risk systems may require more frequent review.

18.4 Review Outcomes

A periodic review may result in:
  • continued operation;
  • updated controls;
  • updated risk assessment;
  • updated governance classification;
  • additional assurance;
  • restricted operation;
  • redesign;
  • replacement; or
  • retirement.

19. Stage 15 — Retirement and Decommissioning

19.1 Purpose

The Retirement and Decommissioning stage governs the controlled withdrawal of an AI system from operational use.

19.2 Retirement Triggers

Retirement may occur when:
  • the business purpose is no longer required;
  • the system is replaced;
  • risks become unacceptable;
  • the system becomes obsolete;
  • the provider is discontinued;
  • the system no longer meets requirements;
  • maintenance becomes impractical; or
  • organizational strategy changes.

19.3 Retirement Planning

Retirement planning should consider:
  • users;
  • dependencies;
  • data;
  • integrations;
  • models;
  • infrastructure;
  • third-party services;
  • records;
  • contractual obligations;
  • security;
  • privacy; and
  • business continuity.

19.4 Decommissioning Activities

Activities may include:
  • disabling access;
  • removing integrations;
  • terminating services;
  • archiving required records;
  • securely disposing of data where required;
  • revoking credentials;
  • removing infrastructure;
  • updating inventories; and
  • communicating retirement.

20. Stage 16 — Post-Retirement Review

20.1 Purpose

The Post-Retirement Review evaluates the outcomes and lessons from the AI system lifecycle after retirement.

20.2 Review Areas

The review may consider:
  • lifecycle performance;
  • incidents;
  • governance effectiveness;
  • control effectiveness;
  • risk decisions;
  • stakeholder outcomes;
  • lessons learned;
  • technical lessons;
  • operational lessons; and
  • opportunities for improvement.

20.3 Knowledge Retention

Relevant information should be retained according to organizational requirements. This may include:
  • system documentation;
  • risk assessments;
  • approval records;
  • incident records;
  • audit evidence;
  • change records;
  • retirement records; and
  • lessons learned.

21. Lifecycle Gates

AIGO may use governance gates to establish decision points between lifecycle stages. Examples include:
  1. Initiation Gate
  2. Assessment Gate
  3. Risk Gate
  4. Design Gate
  5. Development Gate
  6. Validation Gate
  7. Approval Gate
  8. Deployment Gate
  9. Operational Review Gate
  10. Retirement Gate
The number and structure of gates may be adapted according to organizational needs.

22. Lifecycle Entry and Exit Criteria

Each lifecycle stage should define appropriate entry and exit criteria. Entry criteria establish whether sufficient conditions exist to begin a stage. Exit criteria establish whether the required activities and decisions have been completed before proceeding. Where exit criteria are not satisfied, the organization should determine whether to:
  • remediate;
  • defer;
  • restrict;
  • escalate; or
  • terminate the activity.

23. Lifecycle Evidence

Organizations should maintain evidence appropriate to the AI system and its governance requirements. Evidence may include:
  • system records;
  • risk assessments;
  • governance classifications;
  • architecture documentation;
  • technical documentation;
  • test results;
  • approvals;
  • monitoring records;
  • incident records;
  • change records;
  • review records; and
  • retirement records.
Evidence requirements should be proportionate to risk and should support traceability and assurance.

24. Lifecycle Traceability

AIGO lifecycle activities should maintain traceability between governance decisions and the artifacts supporting those decisions. Where practical, organizations should be able to trace: Lifecycle Stage → Activity → Role → Risk → Control → Evidence → Decision Traceability should support:
  • accountability;
  • auditability;
  • assurance;
  • change management; and
  • continuous improvement.

25. Lifecycle Tailoring

Organizations may tailor the AIGO lifecycle according to:
  • organizational size;
  • system complexity;
  • risk;
  • regulatory environment;
  • AI System Profile;
  • development methodology;
  • procurement model;
  • deployment architecture; and
  • operational context.
Tailoring should not remove governance activities that are necessary to manage material risks. Where lifecycle requirements are omitted or combined, the organization should maintain sufficient evidence that equivalent governance outcomes are achieved.

26. Lifecycle for Third-Party AI Systems

Organizations using third-party AI systems or services should apply the AIGO lifecycle to the extent appropriate to their level of control and responsibility. Lifecycle activities may include:
  • provider identification;
  • due diligence;
  • contractual assessment;
  • risk assessment;
  • security and privacy assessment;
  • approval;
  • deployment;
  • monitoring;
  • provider change management;
  • incident management;
  • periodic review; and
  • service termination.
The organization should identify which lifecycle activities are performed internally and which are performed by the third-party provider.

27. Lifecycle for AI Agents

AI agents may require additional lifecycle activities because they may perform actions, invoke tools, interact with external systems, or operate with a degree of autonomy. Organizations should consider:
  • authorized actions;
  • available tools;
  • permissions;
  • external integrations;
  • human oversight;
  • action limits;
  • failure handling;
  • monitoring;
  • escalation; and
  • shutdown mechanisms.
Changes that increase an agent’s autonomy, permissions, or access to external systems should trigger appropriate reassessment.

28. Lifecycle for Agentic Workflows

Agentic workflows may involve multiple AI components operating together to achieve a business objective. The lifecycle should consider:
  • individual AI components;
  • orchestration mechanisms;
  • workflow logic;
  • data flows;
  • tool access;
  • permissions;
  • human intervention;
  • dependencies;
  • monitoring; and
  • failure propagation.
Governance should apply to the overall workflow as well as material individual components.

29. Lifecycle for Generative AI Systems

Generative AI systems may require lifecycle considerations related to:
  • model selection;
  • prompt design;
  • system instructions;
  • retrieval sources;
  • generated content;
  • output validation;
  • content risks;
  • intellectual property;
  • privacy;
  • security;
  • hallucination or inaccurate output;
  • user reliance; and
  • monitoring.
Organizations should define appropriate controls according to the intended use and risk of the system.

30. Lifecycle for AI-Enabled Business Processes

Where AI is embedded within an existing business process, the AIGO lifecycle should consider both the AI system and the overall business process. Assessment should consider:
  • process purpose;
  • decision points;
  • human involvement;
  • process dependencies;
  • downstream impacts;
  • affected stakeholders;
  • control points; and
  • business continuity.
The introduction of AI should not be treated as isolated from the governance of the process in which it operates.

31. Lifecycle for Existing AI Systems

Organizations may discover AI systems that were implemented before the AIGO governance lifecycle was established. Such systems should be brought into the lifecycle through an appropriate retrospective assessment. The retrospective process may include:
  1. identifying the system;
  2. identifying the accountable owner;
  3. documenting intended use;
  4. determining governance classification;
  5. performing risk assessment;
  6. identifying governance gaps;
  7. implementing required controls;
  8. obtaining required approval; and
  9. establishing ongoing monitoring.
The organization should prioritize remediation according to risk.

32. Lifecycle for Unauthorized AI Use

Organizations should establish mechanisms for identifying and responding to AI systems or services that are being used without appropriate authorization. Unauthorized use may include:
  • unapproved AI services;
  • unauthorized models;
  • unapproved AI applications;
  • unauthorized data processing;
  • unapproved AI agents;
  • personal use of organizational data with external AI services; or
  • AI functionality introduced without governance review.
The response should be proportionate to the risk and may include:
  • investigation;
  • containment;
  • removal;
  • risk assessment;
  • remediation;
  • user education; and
  • formalization through the AIGO lifecycle where appropriate.

33. Lifecycle Risk Proportionality

AIGO lifecycle requirements should be proportionate to the potential risk and impact of the AI system. Higher-risk systems may require:
  • additional assessment;
  • more detailed documentation;
  • stronger controls;
  • independent assurance;
  • increased monitoring;
  • additional human oversight;
  • more frequent review; and
  • higher approval authority.
Lower-risk systems may use simplified lifecycle procedures where appropriate. Risk proportionality should not be interpreted as eliminating accountability.

34. Lifecycle Exceptions

Organizations may establish controlled exceptions to standard lifecycle requirements where justified. An exception should:
  • identify the requirement being excepted;
  • document the reason;
  • assess associated risk;
  • identify compensating measures where appropriate;
  • identify an authorized approver;
  • define any conditions or limitations; and
  • establish an expiration or review date where appropriate.
Exceptions should not become a substitute for appropriate governance.

35. Lifecycle Escalation

Lifecycle issues should be escalated when they exceed defined authority, risk tolerance, or governance thresholds. Escalation may be required when:
  • material risks cannot be adequately mitigated;
  • required evidence is unavailable;
  • approval authority is unclear;
  • significant control deficiencies remain;
  • system behavior materially differs from expectations;
  • an incident occurs;
  • a significant regulatory concern is identified; or
  • lifecycle requirements cannot be satisfied.
Escalation paths should be defined in applicable governance procedures.

36. Lifecycle Decision Records

Material lifecycle decisions should be documented. Decision records may include:
  • decision date;
  • decision maker;
  • decision scope;
  • information considered;
  • relevant risks;
  • controls considered;
  • conditions;
  • rationale; and
  • resulting actions.
Decision records should be retained according to organizational requirements.

37. Lifecycle Monitoring

Organizations should monitor the effectiveness of their AI governance lifecycle itself. Lifecycle monitoring may consider:
  • stage completion;
  • approval delays;
  • unresolved risks;
  • control deficiencies;
  • incidents;
  • exceptions;
  • recurring issues;
  • assessment quality;
  • evidence completeness; and
  • governance outcomes.
Lifecycle performance information should be used to identify opportunities for improvement.

38. Lifecycle Metrics

Organizations may establish metrics and indicators for lifecycle governance. Examples include:
  • number of AI systems registered;
  • percentage of systems with assigned owners;
  • percentage of systems with completed risk assessments;
  • percentage of systems with required approvals;
  • number of overdue reviews;
  • number of open governance issues;
  • number of exceptions;
  • incident frequency;
  • control effectiveness; and
  • retirement completion.
Metrics should be interpreted in context and should not be used as the sole measure of governance effectiveness.

39. Lifecycle Documentation

Organizations should maintain documentation appropriate to the lifecycle stage and risk of each AI system. Documentation may include:
  • initiation records;
  • system descriptions;
  • intended use;
  • risk assessments;
  • governance classifications;
  • design documentation;
  • testing evidence;
  • approval records;
  • operational documentation;
  • monitoring records;
  • change records;
  • incident records;
  • periodic reviews; and
  • retirement records.
Documentation should be maintained in a manner that supports traceability and accessibility to authorized stakeholders.

40. Lifecycle Records and Retention

Lifecycle records should be retained according to applicable organizational, legal, regulatory, contractual, and information-management requirements. Retention requirements should consider:
  • record type;
  • business importance;
  • risk;
  • regulatory requirements;
  • contractual requirements;
  • privacy;
  • security; and
  • future assurance needs.
Records should be protected against unauthorized alteration or loss.

41. Lifecycle Roles and Accountability

Each lifecycle stage should have clearly defined responsibility and accountability. Roles identified in the AIGO Governance Roles and Accountability model may participate at different stages according to their responsibilities. Organizations should be able to determine, where appropriate:
  • who performs the activity;
  • who is accountable;
  • who approves the decision;
  • who provides specialist advice;
  • who must be consulted; and
  • who must be informed.

42. Lifecycle and Controls

AIGO lifecycle activities should be connected to applicable controls. Controls within 07-controls may apply to:
  • one lifecycle stage;
  • multiple lifecycle stages;
  • the entire lifecycle; or
  • specific AI System Profiles.
Organizations should establish traceability where practical between lifecycle activities and applicable controls.

43. Lifecycle and Risk Management

The lifecycle should integrate with the AIGO risk model defined in 06-risk. Risk assessment should not be limited to a single point in the lifecycle. Risk should be reviewed when:
  • the system changes;
  • new information becomes available;
  • incidents occur;
  • the operating environment changes;
  • new stakeholders are affected;
  • applicable requirements change; or
  • the system reaches a new lifecycle stage requiring reassessment.

44. Lifecycle and Maturity

Organizations may assess the maturity of their AI lifecycle governance. Maturity considerations may include:
  • lifecycle definition;
  • process consistency;
  • role clarity;
  • risk integration;
  • evidence;
  • automation;
  • monitoring;
  • assurance; and
  • continuous improvement.
Maturity assessment should focus on governance capability and effectiveness rather than simply the existence of documented processes.

45. Lifecycle Continuous Improvement

The AIGO lifecycle should support continuous improvement. Organizations should use information from:
  • incidents;
  • audits;
  • assessments;
  • monitoring;
  • user feedback;
  • stakeholder feedback;
  • regulatory developments;
  • technology changes;
  • lessons learned; and
  • governance metrics
to improve lifecycle processes and controls.

46. Lifecycle Integration with Organizational Governance

AIGO should integrate with existing organizational governance processes where appropriate. Relevant processes may include:
  • enterprise risk management;
  • information security management;
  • privacy management;
  • data governance;
  • software development;
  • procurement;
  • change management;
  • business continuity;
  • internal audit;
  • compliance; and
  • quality management.
AIGO does not require organizations to create duplicate governance processes where existing processes can achieve equivalent outcomes.

47. Lifecycle Integration with External Requirements

Organizations should consider applicable external requirements throughout the AI lifecycle. These may include:
  • laws;
  • regulations;
  • contractual obligations;
  • industry standards;
  • customer requirements;
  • professional requirements; and
  • organizational commitments.
The applicable requirements should be identified according to the organization’s jurisdiction, activities, and risk profile.

48. Lifecycle Traceability Model

The AIGO lifecycle should support traceability across governance artifacts. A representative traceability relationship is: AI System → Lifecycle Stage → Activity → Role → Risk → Control → Evidence → Decision → Outcome This relationship supports:
  • accountability;
  • governance transparency;
  • auditability;
  • assurance;
  • change management; and
  • continuous improvement.

49. Lifecycle Completion

An AI system should not be considered fully governed solely because it has passed an initial approval process. Governance should continue through:
  • operation;
  • monitoring;
  • change;
  • reassessment;
  • incident management;
  • periodic review; and
  • retirement.
Lifecycle completion occurs when the system has been appropriately retired or otherwise transitioned out of the organization’s governance scope.

50. Document Status

Document: AIGO AI Governance Lifecycle Version: 0.1 Status: Draft Working Name: AIGO Full Name: AI Governance Operating Framework Document Type: Framework Lifecycle Identifier Prefix: AIGO-LC This document defines the foundational AI governance lifecycle for the AIGO framework. Organizations may adapt the lifecycle according to their organizational structure, AI systems, risk profile, regulatory environment, and governance maturity while maintaining appropriate accountability, risk management, control, evidence, and assurance.