AIGO — AI Approval Example
AIGO — AI Governance Operating Framework
Version: 0.1 Status: Draft Working Name: AIGO Full Name: AI Governance Operating Framework Document Identifier:AIGO-EXAMPLE-006
Document Type: Implementation Example
Example Type: AI System Approval
1. Purpose
This document provides an illustrative example of how an organization can apply the AIGO AI Approval process before an AI system is deployed, materially changed, continued, suspended, or retired. The example demonstrates how approval can integrate:- AI system registration;
- AI classification;
- risk assessment;
- control assessment;
- evidence review;
- human oversight;
- security and privacy review;
- operational readiness;
- residual-risk evaluation;
- management decision;
- approval conditions;
- post-approval monitoring;
- change management.
2. Example Organization
For this example, the organization is ExampleCorp, a fictional organization implementing AIGO. The organization intends to deploy an AI-enabled recruitment-support system.3. AI System
System Name: Candidate Assessment Assistant AI System ID:AI-HR-001
Business Function: Human Resources
System Owner: HR AI System Owner
Business Owner: HR Director
Model Owner: AI/ML Engineering Lead
Risk Owner: Enterprise Risk Manager
Classification: Class 3 — Enhanced Governance
Lifecycle Stage: Deployment Readiness
4. Approval Objective
The approval process determines whether the AI system is sufficiently governed and ready to:- deploy;
- continue operating;
- undergo a material change;
- resume operation following suspension.
5. Approval Principle
AIGO treats approval as a governance decision. Approval should consider:- intended purpose;
- classification;
- applicable requirements;
- identified risks;
- implemented controls;
- control effectiveness;
- evidence quality;
- residual risk;
- operational readiness;
- human oversight;
- monitoring readiness;
- incident readiness;
- change-management readiness.
6. Approval Lifecycle
7. Approval Trigger
ExampleCorp initiates approval because the Candidate Assessment Assistant has reached the deployment-readiness stage. The approval cannot proceed until the required governance records have been assembled.8. Approval Package
The approval package contains:9. System Registration Review
The approval authority first verifies that the system is registered. Required information includes:- system identifier;
- system owner;
- business owner;
- intended purpose;
- users;
- affected stakeholders;
- lifecycle status;
- classification;
- suppliers;
- dependencies.
10. Intended Purpose
The Candidate Assessment Assistant is intended to support recruitment personnel by:- organizing candidate information;
- identifying relevant experience;
- generating structured candidate summaries;
- providing recruitment-support recommendations.
11. Prohibited or Restricted Uses
ExampleCorp establishes that the system must not:- autonomously reject candidates;
- make final hiring decisions;
- generate decisions using prohibited personal characteristics;
- bypass human review;
- be used outside its approved purpose without reassessment;
- be materially changed without change approval.
12. AI Classification
The system is classified as: Class 3 — Enhanced Governance The classification is based on:- use in employment-related processes;
- potential impact on individuals;
- sensitivity of decisions;
- potential fairness risks;
- need for meaningful human oversight.
13. Risk Assessment Summary
The assessment identifies risks including:14. Risk Treatment Summary
Risk treatments include:- human review;
- fairness testing;
- data-quality controls;
- access control;
- privacy controls;
- monitoring;
- change management;
- approval controls;
- incident management;
- transparency measures.
15. Control Assessment Summary
Overall Control Environment: Partially Effective
16. Approval Readiness Model
17. Human Oversight Review
The approval authority verifies that:- human decision-makers are identified;
- reviewers receive appropriate training;
- AI outputs can be challenged;
- AI outputs can be overridden;
- final decisions remain under authorized human responsibility;
- automation bias is addressed.
18. Human Override
The system must provide an explicit mechanism for authorized personnel to reject or override AI recommendations. Overrides should be logged where required. Example:19. Privacy Review
The privacy review verifies:- data categories;
- purpose limitation;
- access restrictions;
- retention;
- data minimization;
- authorized processing;
- privacy risks;
- incident procedures.
20. Security Review
The security review considers:- authentication;
- authorization;
- privileged access;
- logging;
- vulnerability management;
- secure configuration;
- supplier dependencies;
- incident response.
21. Monitoring Readiness
The system has defined monitoring indicators for:- model performance;
- error rates;
- fairness indicators;
- override frequency;
- incidents;
- complaints;
- drift;
- control failures.
22. Incident Readiness
ExampleCorp has an AI incident procedure covering:- incident identification;
- reporting;
- classification;
- escalation;
- containment;
- investigation;
- corrective action;
- recovery;
- lessons learned.
23. Change Management Readiness
The system is subject to controlled change management. Material changes require:- change identification;
- impact assessment;
- risk reassessment;
- control reassessment;
- testing;
- approval;
- documentation.
24. Evidence Package
The approval package contains evidence including:25. Evidence Sufficiency
Evidence is evaluated for:- completeness;
- authenticity;
- relevance;
- timeliness;
- traceability;
- integrity.
26. Residual Risk
Following implementation of controls, the residual risks are assessed.27. Residual Risk Decision
The residual risk is not zero. The approval authority determines that the remaining risk is acceptable only subject to defined conditions. This distinction is important:Approval does not mean that all risk has been eliminated.
28. Approval Conditions
The following conditions are imposed:- complete the fairness monitoring cycle;
- finalize monitoring escalation thresholds;
- conduct an AI incident-management tabletop exercise;
- complete the outstanding risk reassessment;
- maintain human review of final recruitment decisions.
29. Approval Decision
Decision: Conditionally Approved Approval Scope: Production deployment within the documented intended purpose. Conditions: All conditions in Section 28 must be tracked to completion. Approval Authority: AI Governance Committee.30. Approval Decision Model
31. Approval Outcomes
AIGO supports four primary outcomes.32. Approval Deferral
Approval should be deferred when:- material evidence is missing;
- risk cannot be adequately assessed;
- mandatory controls are not implemented;
- ownership is unclear;
- human oversight is insufficient;
- required approvals are unavailable.
33. Approval Rejection
Approval may be rejected where:- residual risk is unacceptable;
- critical controls cannot be implemented;
- intended use cannot be governed adequately;
- required oversight cannot be established;
- material risks cannot be treated or accepted;
- the system conflicts with organizational governance requirements.
34. Approval Authority
The approval authority must have sufficient authority to make the decision. ExampleCorp assigns the AI Governance Committee responsibility for Class 3 approval. The committee includes representatives from:- AI governance;
- business ownership;
- risk;
- privacy;
- security;
- legal/compliance where applicable;
- technical/model ownership;
- assurance where appropriate.
35. Segregation of Duties
Where practical:- system owner prepares the approval package;
- risk function evaluates risk;
- control owners provide evidence;
- approval authority makes the decision;
- assurance may independently review the decision.
36. Approval Record
AIGO maintains a formal approval record.37. Approval Conditions Register
38. Approval Evidence Chain
39. Deployment Authorization
Following the conditional approval decision, deployment authorization is limited to the approved scope. The system must not:- expand beyond the approved purpose;
- change classification without review;
- materially change without approval;
- disable required controls;
- remove human oversight.
40. Post-Approval Monitoring
Approval does not terminate governance. After deployment, ExampleCorp monitors:- system performance;
- control effectiveness;
- risk indicators;
- fairness;
- incidents;
- complaints;
- changes;
- residual risk.
41. Approval Review Triggers
Approval must be reconsidered when:- intended purpose changes;
- system functionality materially changes;
- model architecture changes materially;
- data sources change materially;
- risk increases;
- material incidents occur;
- significant control failures occur;
- applicable requirements change;
- monitoring identifies significant deterioration.
42. Triggered Reapproval
43. Approval Suspension
The approval authority may suspend approval where:- critical control failure occurs;
- unacceptable risk emerges;
- serious incident occurs;
- required human oversight becomes unavailable;
- system operates outside approved scope;
- material unauthorized change occurs.
44. Approval Revocation
Approval may be revoked where the organization determines that continued operation is no longer acceptable. Revocation may trigger:- system suspension;
- containment;
- stakeholder notification;
- incident management;
- corrective action;
- retirement assessment.
45. Approval and Risk Acceptance
Approval and risk acceptance are related but distinct decisions. Approval: Authorization to operate within defined conditions. Risk Acceptance: Formal decision to accept identified residual risk. A system may require both.46. Example Risk Acceptance
For the Candidate Assessment Assistant: Risk: Fairness monitoring delay Residual Risk: High Risk Owner: Enterprise Risk Manager Decision: Temporarily accepted subject to corrective action. Expiration: Until the next formal review or earlier if risk conditions change.47. Approval and Control Effectiveness
Approval should consider control effectiveness. A system with multiple ineffective critical controls should normally not receive unconditional approval. The decision should reflect:- control criticality;
- risk severity;
- evidence quality;
- compensating controls;
- management acceptance.
48. Approval and Evidence Quality
Insufficient evidence may prevent approval even when the control is believed to exist. For example:49. Approval and Human Oversight
Human oversight is a key approval consideration for systems where AI outputs may affect people or material decisions. Approval should verify that:- oversight is meaningful;
- reviewers have authority;
- reviewers have sufficient information;
- override mechanisms exist;
- responsibility remains clear.
50. Approval and Continuous Improvement
Approval records should feed continual improvement. Lessons from:- incidents;
- monitoring;
- audits;
- assessments;
- complaints;
- control failures;
- changes
51. Management Reporting
The AI Governance Committee should receive appropriate information regarding:- pending approvals;
- conditional approvals;
- overdue conditions;
- rejected systems;
- suspended systems;
- material residual risks;
- recurring control deficiencies.
52. Approval Status Dashboard
Example:53. Approval Traceability
The approval decision should be traceable to the information on which it was based.54. Minimum Approval Record
An AIGO approval record should contain, as applicable:- AI system identifier;
- system name;
- intended purpose;
- classification;
- owner;
- risk owner;
- approval authority;
- approval type;
- assessment results;
- control status;
- evidence status;
- residual risk;
- conditions;
- decision;
- decision date;
- effective date;
- review date;
- approver;
- approval status.
55. Approval Checklist
- AI system registered
- Intended purpose documented
- Classification completed
- Applicable requirements identified
- Risk assessment completed
- Risk treatment documented
- Controls identified
- Controls assessed
- Evidence reviewed
- Human oversight verified
- Privacy review completed
- Security review completed
- Monitoring plan approved
- Incident process established
- Change process established
- Residual risk assessed
- Conditions documented
- Approval authority identified
- Approval decision recorded
- Post-approval monitoring established
56. Example Approval Meeting
The AI Governance Committee reviews:- system purpose;
- classification;
- risk assessment;
- control assessment;
- open findings;
- evidence;
- residual risk;
- operational readiness;
- approval conditions.
57. Example Approval Minutes
Meeting: AI Governance Committee Subject: Candidate Assessment Assistant Decision: Conditionally Approved Key Conditions:- fairness monitoring;
- monitoring thresholds;
- incident exercise;
- risk reassessment;
- ongoing human oversight.
58. Approval Communication
The approval decision should be communicated to relevant stakeholders. Communication should include:- system;
- approved scope;
- decision;
- conditions;
- restrictions;
- effective date;
- review date;
- responsible owners.
59. Unauthorized Operation
If an AI system is found operating without required approval, the organization should initiate the applicable governance process. Potential actions include:- immediate escalation;
- scope restriction;
- temporary suspension;
- risk assessment;
- control assessment;
- retrospective approval review;
- incident assessment.
60. Approval Exceptions
Exceptions to the normal approval process should be:- explicitly justified;
- documented;
- risk assessed;
- time limited;
- approved by authorized personnel;
- subject to retrospective review.
61. Approval Record Retention
Approval records should be retained according to applicable organizational document-control and retention requirements. The record should remain traceable throughout the AI system lifecycle.62. Approval and Lifecycle
Approval is lifecycle-dependent. A system may require different approval decisions at:- initial deployment;
- material change;
- major model update;
- significant risk change;
- continued operation;
- post-incident recovery;
- retirement.
63. Lifecycle Approval Model
64. Approval Decision Quality
A high-quality approval decision should be:- informed;
- documented;
- evidence-based;
- risk-aware;
- authorized;
- traceable;
- reviewable;
- reversible where necessary.
65. Example Final Approval Statement
ExampleCorp determines that the Candidate Assessment Assistant is conditionally approved for deployment within its documented intended purpose. The approval is subject to the conditions identified in the approval conditions register. The system owner remains accountable for compliance with the approved operating conditions and must escalate material changes, incidents, control failures, or changes in residual risk.66. Relationship to AIGO Procedures
This example should be implemented through the applicable AIGO procedures, particularly:- AI Governance Procedure;
- AI System Registration Procedure;
- AI Classification Procedure;
- AI Risk Assessment Procedure;
- AI Control Assessment Procedure;
- AI Approval Procedure;
- AI Change Management Procedure;
- AI Monitoring Procedure;
- AI Assurance Procedure;
- AI Risk Acceptance Procedure;
- AI Incident Management Procedure;
- AI Retirement Procedure.
67. Relationship to AIGO Controls
The approval process demonstrates how AIGO controls are integrated into a governance decision. Approval should not operate as an isolated administrative step. It should consume outputs from:- registration;
- classification;
- risk management;
- control assessment;
- evidence management;
- monitoring;
- assurance.
68. Relationship to ISO/IEC 42001
The approval process can support an AI management system by providing documented governance decisions, responsibilities, risk considerations, operational controls, evidence, and continual-review mechanisms. Applicable ISO/IEC 42001 requirements should be determined separately by the implementing organization.69. Relationship to NIST AI RMF
The approval process can support activities associated with:70. Key Lessons
70.1 Approval Is a Governance Decision
Approval is not merely a signature.70.2 Approval Requires Evidence
The decision should be supported by documented information.70.3 Approval Does Not Eliminate Risk
Residual risk may remain after approval.70.4 Conditional Approval Is Useful
Conditions allow organizations to manage controlled residual issues while maintaining explicit accountability.70.5 Approval Must Continue Through the Lifecycle
Material changes and significant events may require reassessment or reapproval.70.6 Approval Must Be Traceable
A reviewer should be able to reconstruct why the organization approved, deferred, rejected, or suspended a system.71. Complete Approval Model
72. Document Status
Document: AIGO — AI Approval Example Version: 0.1 Status: Draft Working Name: AIGO Full Name: AI Governance Operating Framework Document Identifier:AIGO-EXAMPLE-006
Document Type: Implementation Example
Example Type: AI System Approval
This document provides an illustrative example of how an AI system approval decision can be performed and documented within the AIGO AI Governance Operating Framework.
73. End of Example Document
AIGO — AI Approval Example Document ID:AIGO-EXAMPLE-006
Version: 0.1
Status: Draft
End of Document