PRGoverning principles
TA-14 AI GOVERNANCE LIBRARY
AI Governance Principles
Explore the foundational principles that shape AI laws, standards, frameworks, management systems, assurance practices, organizational responsibilities, and evidence-bound execution decisions.
GPPrinciple function
PRINCIPLE GOVERNANCE
Principles define the values governance must protect, but evidence determines whether those values were actually upheld.
A principle becomes operational only when it is translated into roles, controls, thresholds, records, review requirements, execution conditions, and preserved outcomes. Without that translation, principle language may guide intent without governing action.
PRINCIPLE CONTROL DESK
Search the governing values that shape trustworthy AI decisions.
Compare principle categories, governing questions, expected obligations, required evidence, related governance areas, and the execution boundaries each principle establishes.
12Principles displayed
0Active filters
0Principles expanded
HA
01
Human AuthorityFoundational
Human Agency and Oversight
AI systems should preserve meaningful human authority, intervention paths, informed review, and accountability for consequential decisions.
Governing question
Can an authorized human understand, challenge, interrupt, override, or refuse the system’s proposed action before harm becomes irreversible?
Operational obligations6 duties
◆Define accountable human decision authorities
◆Preserve intervention and override capability
◆Prevent nominal or ineffective human review
◆Document escalation and exception routes
Expected evidence
Authority Matrix
Human Oversight Plan
Intervention Record
Escalation Log
Reviewer Competence Record
Approval Receipt
Execution boundaryExecution must be held when required human authority is absent, uninformed, conflicted, or unable to intervene.
TS
02
System IntegrityFoundational
Technical Robustness and Safety
AI systems should operate reliably, resist foreseeable failure, remain within approved conditions, and fail safely when boundaries are exceeded.
Governing question
Does the system remain dependable under expected, degraded, adversarial, and out-of-distribution operating conditions?
Operational obligations6 duties
◆Define operational and safety thresholds
◆Test foreseeable misuse and failure modes
◆Implement fallback, rollback, and safe-state controls
◆Monitor reliability, drift, and boundary violations
Expected evidence
Safety Case
Robustness Test Results
Failure Mode Analysis
Rollback Verification
Configuration Record
Revalidation Decision
Execution boundaryExecution must not proceed when the system is outside validated conditions or when safe failure cannot be assured.
DG
03
Data GovernanceFoundational
Privacy and Data Governance
Data should be lawfully obtained, appropriately governed, traceable, secure, proportionate, and limited to legitimate purposes.
Governing question
Can the organization prove where the data came from, why it may be used, how it was transformed, and who remains accountable for it?
Operational obligations6 duties
◆Establish lawful basis and legitimate purpose
◆Preserve provenance and transformation history
◆Minimize collection and retention
◆Control access, sharing, and secondary use
Expected evidence
Data Provenance Record
Lawful Basis Assessment
Data Inventory
Access Control Record
Privacy Impact Assessment
Retention Decision
Execution boundaryExecution must be restricted when data authority, provenance, permitted purpose, integrity, or access control cannot be demonstrated.
TX
04
Information IntegrityFoundational
Transparency and Explainability
Relevant parties should be able to understand the system’s role, limitations, evidence basis, decision path, and material consequences.
Governing question
Is the explanation sufficient for the affected party, operator, reviewer, regulator, or decision authority who must rely on it?
Operational obligations6 duties
◆Disclose AI involvement where material
◆Explain intended purpose and limitations
◆Preserve decision logic and evidence basis
◆Provide audience-appropriate explanations
Expected evidence
System Disclosure
Model Card
Decision Explanation
Limitation Statement
Confidence Record
Challenge Response
Execution boundaryExecution should be held when required parties cannot understand the decision, its basis, or its material limitations.
FN
05
Rights and EquityFoundational
Fairness and Non-Discrimination
AI governance should identify, evaluate, prevent, and remedy unjustified differential treatment, exclusion, and harmful bias.
Governing question
Are materially different outcomes justified by legitimate evidence, or do they reflect avoidable bias, proxy discrimination, or structural exclusion?
Operational obligations6 duties
◆Identify affected groups and protected interests
◆Assess representativeness and data imbalance
◆Test differential outcomes and error rates
◆Document justification for material disparities
Expected evidence
Fairness Assessment
Dataset Representation Review
Disparity Test Results
Justification Record
Remedy Record
Post-Deployment Monitoring
Execution boundaryExecution must be restricted when material disparity lacks lawful, proportionate, evidence-supported justification.
AA
06
Governance AuthorityFoundational
Accountability and Auditability
Roles, authority, evidence, approvals, execution decisions, interventions, and outcomes should be attributable and reviewable.
Governing question
Can every consequential decision be traced to the responsible actors, governing authority, evidence basis, and preserved outcome?
Operational obligations6 duties
◆Assign ownership for systems and decisions
◆Separate review, approval, and execution roles
◆Preserve complete decision and action records
◆Maintain traceable authority and delegation
Expected evidence
Responsibility Matrix
Delegation Record
Decision Log
Execution Receipt
Audit Trail
Corrective Action Record
Execution boundaryExecution must not proceed when responsibility, authority, decision ownership, or record preservation is materially unclear.
EW
07
Public InterestFoundational
Environmental and Social Well-Being
AI systems should account for broader environmental, social, institutional, and public-interest consequences across their lifecycle.
Governing question
What direct, indirect, cumulative, and externalized effects could the system impose on people, communities, institutions, or the environment?
Operational obligations6 duties
◆Assess environmental and social impacts
◆Identify affected communities and stakeholders
◆Evaluate energy, resource, and infrastructure burdens
◆Consider systemic and cumulative consequences
Expected evidence
Impact Assessment
Stakeholder Record
Resource Use Record
Public Interest Analysis
Mitigation Plan
Long-Term Monitoring Record
Execution boundaryExecution should be restricted when foreseeable public or environmental harm is unassessed, unmanaged, or disproportionate.
SR
08
Operational ProtectionFoundational
Security and Resilience
AI systems, data, interfaces, dependencies, and governing records should be protected against unauthorized access, manipulation, disruption, and compromise.
Governing question
Can the system preserve trusted operation and recover safely when attacked, corrupted, interrupted, or deprived of critical dependencies?
Operational obligations6 duties
◆Identify threat actors and attack surfaces
◆Secure models, data, interfaces, and credentials
◆Test adversarial and abuse scenarios
◆Monitor compromise and unauthorized change
Expected evidence
Threat Model
Security Test Results
Access Log
Incident Record
Recovery Verification
Supplier Security Assessment
Execution boundaryExecution must be denied or suspended when system integrity, access control, dependency trust, or recovery capability is compromised.
CR
09
Procedural RightsFoundational
Contestability and Remedy
Affected parties should have accessible routes to question, challenge, correct, appeal, and obtain remedy for consequential AI outcomes.
Governing question
Can a materially affected person meaningfully contest the decision and obtain review by an authorized party with power to change the outcome?
Operational obligations6 duties
◆Provide notice of consequential AI use
◆Establish accessible challenge routes
◆Preserve evidence needed for review
◆Assign independent appeal authority
Expected evidence
Decision Notice
Challenge Submission
Review Record
Appeal Decision
Correction Record
Remedy Closure
Execution boundaryExecution should be held or reversed when required notice, review, correction, appeal, or remedy mechanisms are unavailable.
PP
10
Use GovernanceFoundational
Proportionality and Purpose Limitation
AI use, control intensity, data processing, and intervention should remain proportionate to the declared purpose, risk, and affected interests.
Governing question
Is this use necessary, suitable, bounded, and proportionate to the legitimate objective it claims to serve?
Operational obligations6 duties
◆Declare specific intended purpose
◆Prohibit incompatible secondary use
◆Match controls to risk and consequence
◆Use the least intrusive effective approach
Expected evidence
Purpose Statement
Necessity Assessment
Proportionality Review
Use Restriction Record
Purpose Drift Assessment
Retirement Decision
Execution boundaryExecution must not proceed when the use exceeds declared purpose, necessity, authority, or proportionality.
EC
11
Record IntegrityTA-14 Governing Principle
Evidence Integrity and Continuity
Governance evidence should remain attributable, authentic, complete, temporally relevant, and continuous from reality through outcome.
Governing question
Can the evidence supporting this decision be trusted as a complete and current representation of the governed reality?
Operational obligations6 duties
◆Preserve origin, custody, and transformation history
◆Maintain version and temporal continuity
◆Prevent unauthorized alteration or substitution
◆Identify gaps, conflicts, and stale evidence
Expected evidence
Reality Record
Chain-of-Custody Record
Version History
Continuity Determination
Binding Record
Outcome Evidence
Execution boundaryNo consequential execution should occur when the evidence chain is incomplete, stale, conflicting, unauthenticated, or disconnected from the governed reality.
AE
12
Execution GovernanceTA-14 Governing Principle
Admissible Execution
Execution should occur only when evidence, authority, continuity, applicability, binding, and governing conditions are satisfied.
Governing question
Has the proposed action earned permission to alter reality under the applicable authority, evidence, controls, and conditions?
Operational obligations6 duties
◆Establish admissible truth before intervention
◆Determine applicable authority and requirements
◆Verify continuity of evidence and system state
◆Bind conditions, limitations, and decision authority
Expected evidence
Applicability Determination
Admissibility Decision
Authority Record
Binding Record
Execution Receipt
Outcome Record
Execution boundaryNo admissible evidence. No admissible execution.
FROM PRINCIPLE TO GOVERNED EXECUTION
A principle must be translated into something the organization can prove, enforce, and preserve.
Declare the value or protected interest.
Translate the principle into a governing obligation.
Define the mechanism that enforces the obligation.
Preserve proof that the control exists and operated.
Determine whether the evidence satisfies the requirement.
Permit, hold, deny, or escalate the proposed action.
AEAdmissible execution
PRINCIPLE-TO-EXECUTION BOUNDARY
Principles guide governance. Admissibility determines whether the action may proceed.
Human oversight, fairness, transparency, safety, privacy, accountability, resilience, and public interest are not satisfied by statements of intent alone. Their governing force depends on whether the required evidence exists, whether authority remains valid, whether controls are operating, and whether the proposed execution remains inside the approved conditions.
PRINCIPLE→REQUIREMENT→CONTROL→EVIDENCE→ADMISSIBILITY→EXECUTION→OUTCOME