AI Governance Playbook

Strategic Enabler for the UN Secretariat | GeoNow 2025

1. Mission: Peace Through Scalable Cooperation

Overview

The United Nations was founded on one core principle: peace through cooperation. Our mission has always been to connect nations, enable dialogue, and coordinate solutions that no single state could achieve alone.

Today, that same mission extends into the digital domain. Where once we convened diplomats, we now connect systems of intelligence. Where once we mediated state-to-state conflict, we now mediate between algorithms, data, and platforms.

Core Principle: The UN's strength has never been control—it has always been coordination. In this new era, coordination happens through AI governance: the rules, architectures, and partnerships that allow us to scale AI responsibly across borders.

Governance, in this context, is not bureaucracy. It is the infrastructure of peace—ensuring that artificial intelligence strengthens human collaboration rather than replacing it.

Objectives

  • Align AI adoption with UN values of trust, transparency, and fairness
  • Enable interoperable and safe data sharing across Secretariat entities
  • Build confidence that AI systems serve humanity, not hierarchy

"In the 20th century, we safeguarded peace through diplomacy. In the 21st century, we safeguard peace through digital cooperation, algorithmic accountability, and data ethics."

1A. Governance Context - The 2025 Landscape

The New UN AI Governance Architecture

In August 2025, the UN General Assembly established two landmark mechanisms:

193 Member States Engaged
2 New Governance Bodies
118 Countries Not in Any Initiative

1. The Global Dialogue on AI Governance

  • Inclusive forum for governments and stakeholders
  • Addresses safety, security, trustworthiness challenges
  • Launched September 2025 during UNGA High-Level Week
  • Annual sessions alternating between New York and Geneva

2. The Independent International Scientific Panel on AI

  • First global scientific body on AI assessment
  • ~40 experts appointed for three-year terms
  • Provides evidence-based policy guidance
Reference: UN General Assembly Resolution A/RES/79/325 (August 26, 2025)
UN News | WEF Analysis

UN Secretariat's Role

OICT serves as the operational bridge between high-level commitments and practical implementation across UN entities.

Our Position: "We translate principles into infrastructure—turning governance commitments into deployable systems that serve peace."

Strategic Alignment

Initiative Organization Year
Global Digital Compact UN (Pact for the Future) 2024
Global AI Governance Action Plan World AI Conference 2025
Hiroshima AI Process G7 2023
UNESCO Ethics Recommendation UNESCO (194 members) 2021
OECD AI Principles OECD, adopted by G20 2019

2. The Strategic Frontier - What AI Represents

AI as a Capability Infrastructure

Nations now compete for informational sovereignty: the ability to forecast, decide, act, and communicate at unprecedented scale and speed.

Eight Strategic Capabilities of AI

Capability Description National Advantage Governance Implication
1. Predictive Sovereignty Forecasting crises before escalation Early warning enables preemptive action Model transparency, validation, bias audits
2. Cognitive Superiority Faster decision cycles Minutes vs. days in crisis response Human oversight, explainability standards
3. Narrative Control Shaping information flows via generative AI Controls perception and messaging Provenance tracking, authenticity controls
4. Knowledge Aggregation Synthesizing global data into intelligence Transforms fragmented info to insight Data-sharing ethics, access equity
5. Algorithmic Economic Power Optimizing production and logistics Efficiency and cost reduction Access fairness, antitrust considerations
6. Cyber-Physical Autonomy AI in drones, grids, transportation Autonomous operations at scale Safety protocols, human-in-loop governance
7. Geospatial Intelligence AI-enhanced satellite/drone monitoring Unprecedented situational awareness Cross-border governance, privacy protection
8. Agentic AI Self-coordinating autonomous systems Future AI-to-AI diplomacy Trust protocols, autonomous audit mechanisms

Impact Quantification

Without Governance

  • Predictive bias: $100M+ misallocation
  • Unverified geospatial AI: false security alerts
  • Algorithmic power: deepened inequality

With Governance

  • Crisis response: 40-60% faster
  • Decision accuracy: +25-35%
  • Trust in AI: +50-70%

Strategic Imperative: The UN Secretariat ensures these capabilities serve peace, equity, and cooperation—not dominance.

3. Framework - Stackable Governance Architecture

Our Model

We govern AI through stackable architecture: independent modules that interlock, scale, and adapt.

Key Principle: Deploy interoperable modules that fit multiple missions, entities, and regions—not one monolithic system.

The Six-Layer Governance Stack

Layer 6: Stakeholder Communication
Model cards, transparency reports, partner dialogue, incident disclosure
Layer 5: Performance Measurement
Fairness metrics, robustness measures, explainability scores
Layer 4: Software & Model Oversight
Model registry, versioning, audit trails, transparency docs
Layer 3: Infrastructure Governance
Data classification, DLP, sensitivity labels, access control
Layer 2: Risk Management
Risk registry, bias detection, robustness testing
Layer 1: Compliance & Ethics
EU AI Act, NIST RMF, ISO 42001, OECD alignment

Framework Alignment

Module EU AI Act NIST RMF ISO 42001
Compliance & Ethics Articles 5-7 Govern Clause 4.4, 5
Risk Management Articles 9-15 Map, Measure, Manage Clause 6, 8
Infrastructure Article 10 Data Governance Clause 7.5
Model Oversight Articles 11-12 Measure Clause 8.2
Performance Article 15 Measure Clause 9
Communication Article 13 Culture Clause 7.4
References: EU AI Act | NIST RMF | ISO 42001

Implementation Path

  • Minimum Viable: Layers 1-3 (start here)
  • 6 Months: Add Layers 4-5
  • 12 Months: Full stack with Layer 6

4. Capabilities - How We Govern

Three Governance Pillars

Pillar 1 Accountability
Pillar 2 Policies
Pillar 3 Partnerships

Pillar 1: Accountability

Clear ownership, traceability, and responsibility for every AI system.

  • AI System Registry: Unique identifier, owner, documentation
  • Stewardship Model: Named individuals responsible
  • Audit Trails: Complete logging of decisions and changes
  • Incident Response: Clear escalation paths

Pillar 2: Policies

Codified standards and automated enforcement.

  • Technical: Data access, model training, deployment standards
  • Ethical: Fairness, bias mitigation, human rights
  • Operational: Usage guidelines, incident reporting
  • Vendor: Third-party AI evaluation, contract requirements

Pillar 3: Partnerships

Collaborative relationships enabling shared governance.

Partner Type Examples Contribution
Hyperscalers Microsoft, Google, AWS Co-development of controls, early access
Academia Research institutions Independent validation, innovation
Member States UN member countries Joint testbeds, capacity building
Civil Society NGOs, advocacy groups Ethics review, accountability

Platform Orchestrator Model

We no longer just set policy—we orchestrate a global governance platform.

  1. Convene diverse stakeholders
  2. Standardize frameworks and protocols
  3. Validate AI systems independently
  4. Enable tools, training, resources
  5. Monitor governance effectiveness
  6. Evolve continuously

4A. Decision Framework - When to Apply Governance

Risk Assessment Dimensions

Dimension Range Governance Impact
Impact Scope Individual → Global Wider scope = higher level
Decision Consequence Informational → Automated More automation = more oversight
Data Sensitivity Public → Restricted Higher sensitivity = stricter controls
Vulnerability Context General → Conflict Zone Vulnerable populations = max safeguards

Five Governance Levels

10-15 pts Level 1 Minimal
16-25 pts Level 2 Limited
26-35 pts Level 3 Moderate
36-45 pts Level 4 High
46-50 pts Level 5 Critical

10-Question Rapid Assessment

Score each question 1-5 points (1=lowest risk, 5=highest):

  1. How many people affected? (1-10 people = 1pt → 10K+ = 5pts)
  2. Data type? (Public = 1pt → Special category = 5pts)
  3. Automation level? (Info only = 1pt → Fully automated = 5pts)
  4. Reversibility? (Instant = 1pt → Irreversible = 5pts)
  5. Vulnerable populations? (None = 1pt → Life-sustaining/crisis = 5pts)
  6. Explainability? (Fully transparent = 1pt → Black box = 5pts)
  7. Bias potential? (None = 1pt → Extreme risk = 5pts)
  8. Cybersecurity risk? (Public/non-sensitive = 1pt → National security = 5pts)
  9. Operational criticality? (Nice to have = 1pt → Mission-critical = 5pts)
  10. Political sensitivity? (None = 1pt → Diplomatic implications = 5pts)

Pre-Deployment Checklist

Before deployment, confirm:
  • ☐ Risk assessment completed
  • ☐ AI Steward assigned
  • ☐ Required governance modules implemented
  • ☐ Data sensitivity classified
  • ☐ Model documentation completed
  • ☐ Bias testing conducted
  • ☐ Cybersecurity review passed
  • ☐ User training prepared
  • ☐ Incident response plan documented
  • ☐ Monitoring configured

5. Governing Entity - OICT Policy, Strategy & Governance Division

About the Division

AI governance is led by the Policy, Strategy & Governance Division within the Office of Information and Communications Technology (OICT).

Mission: Enable UN Secretariat entities to leverage AI safely, ethically, and effectively in service of peace, development, and human rights.

Global Footprint

Location Role Focus Areas
New York, USA Headquarters Strategic oversight, policy development, partnership management
Bangkok, Thailand Asia-Pacific Hub Regional operations, multilingual AI development
Valencia, Spain Global Service Centre European operations, GDPR compliance, data center services
Brindisi, Italy Global Service Centre Field support operations, peacekeeping mission ICT, logistics support

Our Data Landscape

50+ Data Repositories
18-25 Secretariat Entities
2.5-3 PB Archive Size
90% Multilingual Coverage
80%+ Internal Enterprise Data

Specialized Datasets

  • Humanitarian Data Exchange (HDX) - Crisis and development data
  • UNDRR Risk Data - Disaster risk and resilience information
  • Geospatial Data - Satellite imagery from UN-SPIDER and UNOOSA
  • Peacekeeping Mission Data - Operational field mission data
  • Diplomatic Archives - Historical treaties, resolutions, negotiations

Service Lines

Service Description
Enterprise Solutions M365 deployment (including Copilot), ERP, business intelligence
ICT Security CSOC, identity management, DLP, incident response
Field Support Peacekeeping ICT, humanitarian emergency tech, remote deployment
Business Relationship Liaison with departments, requirements gathering, change management

6. Governance Agenda - Strategic Questions

Key Strategic Questions

Question 1: Who governs what Copilot can learn?

Microsoft 365 Copilot now has access to petabytes of UN data. Without governance, we risk:

  • Uncontrolled training on sensitive diplomatic communications
  • Exposure of confidential information through AI-generated responses
  • Loss of data sovereignty to commercial AI systems

Question 2: Can we innovate safely?

Our highest-value enterprise data is also our highest risk:

  • Balancing AI innovation with security requirements
  • Enabling productivity gains without compromising safety
  • Managing vendor dependencies while maintaining control

Question 3: Will AI serve all UN languages equally?

The UN holds the world's largest multilingual diplomatic corpus:

  • Commercial AI models often underperform in non-English languages
  • Risk of perpetuating English-language dominance
  • Need for specialized models trained on diplomatic language across all six official UN languages

Question 4: How is UN data represented in external models?

When commercial AI references "UN data," our credibility is at stake:

  • Ensuring accurate representation of UN positions and data
  • Preventing misattribution or misinterpretation
  • Maintaining authority over official UN information

These are not hypothetical issues—they define the Secretariat's governance roadmap.

7. Challenges - Current Realities

Six Governance Roadblocks

1. Loss of Data Control

UN data used in commercial AI training without consent or oversight.

  • Public UN datasets scraped for model training
  • No attribution or accuracy verification
  • Risk of perpetuating outdated or incorrect information

2. Vendor Dependence

Governance tied to vendor roadmaps and privacy policies.

  • Limited control over feature deployment timing
  • Dependency on vendor governance capabilities
  • Risk of vendor lock-in affecting governance portability

3. Data Leakage Risk

Using third-party AI services may inadvertently train external models.

  • Unclear data handling practices by AI providers
  • Risk of sensitive information exposure through AI outputs
  • Need for contractual safeguards and technical controls

4. Federated Ownership Constraints

Fragmented data ownership prevents full AI access and deployment.

  • 18-25 entities with different priorities and systems
  • No single "owner" requires coordination mechanisms
  • Varying technical capacity across entities

5. Language Equity Gaps

AI models underperform in non-English and low-resource contexts.

  • Commercial models optimized for English
  • Limited training data in other UN official languages
  • Risk of perpetuating language-based inequities

6. Ethical Integration at Speed

Humanitarian missions provide ideal settings, but require governance frameworks to match pace and scale.

  • Rapid deployment needs vs. thorough governance
  • Vulnerable population protection requirements
  • Balancing innovation with ethical safeguards

Mitigation Approach: Sociotechnical solutions combining policy, risk assessment, ethics frameworks, and technical standards—recognizing the landscape evolves rapidly.

8. Countermeasures - How We're Responding

Legal & Policy Actions

  • UN-specific AI terms with commercial partners
  • Data residency controls and opt-out mechanisms for AI training
  • UN AI data licensing framework and attribution requirements
  • robots.txt / HTTP headers and API rate enforcement for public datasets

Technical Safeguards

  • Tiered AI access based on data sensitivity
  • Secure AI sandboxes for high-risk work
  • Audit trails and explainability standards
  • AI-specific data access governance with automated controls

Language & Equity Initiatives

  • Multilingual AI training across all six official UN languages (Arabic, Chinese, English, French, Russian, Spanish)
  • Language-equity requirements in vendor contracts
  • Diplomatic domain-specific models trained on UN corpus
  • Shared datasets with developing countries for capacity building

Procurement & Vendor Oversight

  • Standard AI contractual clauses in every tech agreement
  • Mandatory pre-deployment evaluation of AI features
  • Contractual exit/portability rights for data and models
  • Vendor performance monitoring against governance standards

Organizational Evolution

  • Global expansion of governance partnerships
  • Personnel relocation to operational hubs for real-time oversight
  • Cross-functional teams spanning legal, technical, and policy domains
  • Continuous learning from operational experience

Result: A resilient governance ecosystem—adaptive, accountable, and global.

9. Implementation Pathway - Governance Maturity Model

Five-Level Maturity Model

Level Focus Area Outcome Timeframe
Level 1 Policy & Compliance Baseline AI controls established 0-6 months
Level 2 Risk & Data Management Standard AI safeguards operational 6-12 months
Level 3 Cross-Entity Governance Shared frameworks & data standards 12-18 months
Level 4 Global Interoperability Multi-jurisdictional AI scaling 18-30 months
Level 5 Ethical Intelligence Continuous trust governance & global alignment 30+ months

Implementation Phases

Phase 1: Foundation (Months 1-6)

  • Establish governance structure and leadership
  • Deploy Layers 1-3 of governance stack
  • Create AI system registry
  • Develop initial policies and procedures
  • Begin stakeholder engagement

Phase 2: Operationalization (Months 6-12)

  • Add Layers 4-5 to governance stack
  • Implement automated controls and monitoring
  • Launch AI Steward network across entities
  • Begin multilingual AI initiatives
  • Establish vendor governance protocols

Phase 3: Scaling (Months 12-24)

  • Complete full governance stack (Layer 6)
  • Expand to all Secretariat entities
  • Launch external partnerships and testbeds
  • Publish transparency reports
  • Begin capacity building for Member States

Phase 4: Leadership (Months 24+)

  • Drive global governance standards
  • Enable AI-powered multilateralism
  • Share best practices internationally
  • Continuous improvement and innovation

Success Metrics: % systems governed, incident response time, stakeholder trust scores, language equity measures, vendor compliance rate

10. Conclusion - Governance as the Infrastructure of Peace

Core Message

AI governance is not a barrier to innovation—it is the foundation that makes innovation sustainable.

Our Mission Continues

In the 20th century, the United Nations safeguarded peace through diplomacy.
In the 21st century, we safeguard peace through digital cooperation, algorithmic accountability, and data ethics.

We are not governing code.
We are governing power, knowledge, and trust.

Call to Action

"Every dataset, every model, and every decision can contribute to peace—if it is governed well."

This playbook invites all Secretariat entities and partners to join a coordinated effort: building a trustworthy, multilingual, globally interoperable AI future.

Next Steps

  1. Assess your current AI governance maturity using the framework in Section 9
  2. Identify governance gaps using the decision framework in Section 4A
  3. Engage with OICT to access governance tools and resources
  4. Join UN80 working groups to contribute to global AI governance
  5. Share your learnings to strengthen the collective governance ecosystem

Three Horizons

Horizon Timeframe Focus Goals
Horizon 1 2025-2026 Stabilize Establish baseline governance, secure current systems
Horizon 2 2026-2028 Scale Expand across Secretariat, deepen partnerships
Horizon 3 2028-2030 Lead Drive global standards, enable AI-powered multilateralism

"AI governance is the infrastructure of peace."

Additional Resources

1/12