Effective AI safety governance requires solving challenges across multiple dimensions simultaneously: designing legitimate institutions, building trustworthy actors, establishing justified authority, enabling democratic participation, navigating sovereignty constraints, and distributing costs fairly. Previous frameworks addressed these piecemeal. This paper integrates all six pillars—legitimacy, trust, authority, democracy, sovereignty, and distributive justice—into a unified theory that shows how they interconnect and what governance design must accomplish.
The Six Pillars
AI safety governance rests on six interconnected pillars:
- Legitimacy: the right to govern
- Trust: reliance on actors
- Authority: power to create obligations
- Democracy: participation in decisions
- Sovereignty: supreme authority within territories
- Distributive Justice: fair allocation of costs and benefits
Each pillar has been analyzed separately. But governance must address all simultaneously—and the pillars interact in complex ways.
How the Pillars Connect
The Domestic Stack
Within a single state, the pillars form a hierarchy:
- Democratic Foundation: citizens authorize governance through political participation
- Legitimate Institutions: this authorization creates legitimate governing bodies
- Trustworthy Actors: legitimate institutions cultivate competent, committed actors
- Justified Authority: legitimate, trustworthy institutions can claim authority to bind
- Fair Distribution: authority is exercised to distribute benefits and burdens justly
Each layer depends on those below. Authority without legitimacy is mere coercion. Distribution without authority lacks enforcement. Trust without accountability enables capture.
The International Challenge
But AI safety governance is not domestic—it's global. This introduces sovereignty and international distributive justice as complicating factors:
- No global sovereign: no single authority can govern AI globally
- Sovereignty constraints: states resist constraints on their autonomy
- Distributive disagreement: no consensus on fair distribution of costs
- Variable participation: not all states will participate equally
The Connection Pattern
The six pillars interconnect systematically:
Legitimacy ↔ Trust
- Legitimate institutions enable trust by creating accountability
- Trust enables legitimacy by generating voluntary compliance
- They can diverge: legitimate but distrusted; trusted but illegitimate
- Effective governance needs both: legitimacy that is trustworthy
Legitimacy + Trust → Authority
- Only legitimate, trustworthy institutions can claim authority
- Authority without legitimacy is coercion; without trust, it's unstable
- Authority must address the particularity problem: why this institution?
Democracy → Legitimacy + Authority
- Democratic participation provides consent-based legitimacy
- Democratic authorization grounds authority claims
- But expertise challenges democracy: tension with technical complexity
Sovereignty ↔ All Domestic Pillars
- States claim supreme authority within territories
- This sovereignty can conflict with external authority claims
- International governance must work with, not against, sovereignty
- Historical precedents show sovereignty can be pooled or circumscribed
Distributive Justice ↔ All Pillars
- How costs/benefits are distributed affects legitimacy
- Unfair distribution undermines trust in institutions
- Authority that distributes unfairly loses legitimacy
- Democratic decisions about distribution face disagreement
- Sovereignty allows states to reject distributions they find unfair
The Complete Framework
Putting it together, AI safety governance must:
Level 1: Establish Democratic Foundations
- Domestic: democratic authorization of national AI policies
- International: state consent to international agreements
- Challenge: democratic deficit in international institutions
Level 2: Build Legitimate Institutions
- National regulators: authorized through domestic political processes
- International bodies: authorized through treaty consent
- Challenge: establishing legitimacy across diverse political systems
Level 3: Cultivate Trustworthy Actors
- Competence: technical expertise in AI safety
- Commitment: demonstrated dedication to safety over other interests
- Challenge: industry capture, political pressure, expertise gaps
Level 4: Establish Justified Authority
- Scope: clear domain of authority (what AI safety covers)
- Limits: what authority cannot require (moral limits)
- Particularity: why this institution deserves compliance
- Challenge: justifying authority to those who didn't consent
Level 5: Navigate Sovereignty
- Voluntary pooling: states accept constraints for benefits of coordination
- R2P logic: sovereignty as responsibility—failure to protect may justify intervention
- Variable geometry: different participation levels for different states
- Challenge: sovereignty reassertion, defection, free-riding
Level 6: Distribute Fairly
- Minimum duties: assistance for capacity building (universally accepted)
- Proportionality: contributions proportional to AI development, risk creation
- Negotiated distribution: what states accept through fair negotiation
- Challenge: theoretical disagreement about fair distribution
Design Principles
1. Build All Pillars Simultaneously
Don't focus on one pillar at others' expense. Governance needs:
- Democratic authorization (not just expert rule)
- Legitimate institutions (not just effective ones)
- Trustworthy actors (not just powerful ones)
- Justified authority (not just coercive capacity)
- Sovereignty-compatible design (not utopian global government)
- Fair distribution (not winner-take-all)
2. Design for Incompleteness
No governance will achieve full achievement on all pillars. Design for:
- Partial legitimacy: institutions accepted by most, not all
- Limited trust: verification mechanisms when trust is incomplete
- Contested authority: appeal processes for those who reject authority
- Imperfect democracy: expert checks on democratic decisions
- Variable sovereignty: some states participate fully, others partially
- Disputed distribution: mechanisms for addressing fairness complaints
3. Plan for Vicious Cycles
Governance can spiral downward:
- Illegitimacy → distrust → non-compliance → coercion → more illegitimacy
- Unfair distribution → resentment → withdrawal → coordination failure → worse outcomes
- Sovereignty assertion → treaty withdrawal → regime collapse → race dynamics
Counter with: transparency, appeal mechanisms, sunset provisions, distributed authority, fairness reviews.
4. Cultivate Virtuous Cycles
Governance can spiral upward:
- Legitimacy → trust → voluntary compliance → effectiveness → more legitimacy
- Fair distribution → participation → coordination → benefits → support for more distribution
- Sovereignty pooling → benefits → deeper pooling → more benefits
5. Use Hybrid Structures
Combine expertise and democracy:
- Expert analysis with democratic decision
- Democratic values with expert implementation
- Deliberative processes bringing citizens and experts together
Combine national and international:
- National implementation of international standards
- International coordination of national policies
- Variable geometry with core and associate members
6. Address the Particularity Problem
Why this institution? Answer with:
- Proper authorization (legitimate creation)
- Comparative advantage (better than alternatives)
- Practical necessity (only viable option)
- Fairness (others participating, so should you)
7. Plan for Persistent Disagreement
Deep disagreement about AI risks, fair distribution, and appropriate governance will persist:
- Allow jurisdictional variation when possible
- Build consensus incrementally
- Create deadlock-breaking procedures
- Accept that some governance will lack full legitimacy
Application to AI Safety Governance
National Regulation
Challenge: establishing legitimacy and authority over AI companies
Six-pillar approach:
- Democratic: public input on AI policy priorities
- Legitimate: authorized through normal political processes
- Trustworthy: competent regulators, transparent processes
- Authoritative: clear mandate, justified scope
- Sovereign: within national jurisdiction
- Distributive: fair allocation of compliance costs
International Cooperation
Challenge: building governance without global sovereign
Six-pillar approach:
- Democratic: state consent through treaty processes
- Legitimate: representing will of participating peoples
- Trustworthy: demonstrated competence, verification mechanisms
- Authoritative: address why this institution, not others
- Sovereignty: voluntary pooling, variable participation
- Distributive: fair cost distribution addressing free-riding
Technical Standards
Challenge: expertise-democracy tension
Six-pillar approach:
- Democratic: stakeholder input, not just experts
- Legitimate: proper authorization, accountability
- Trustworthy: technical competence, independence
- Authoritative: limited to technical domain, not value choices
- Sovereignty: national adoption decisions retained
- Distributive: equitable access to standards and expertise
The Governance Stack: Complete Version
Effective AI safety governance requires:
- Foundation: Democratic/Consent Authorization
Public authorization domestically; state consent internationally - Structure: Legitimate Institutions
Institutions justified through proper processes - People: Trustworthy Actors
Competent, committed actors with track records - Rules: Justified Authority
Clear scope, moral limits, particularity addressed - Scope: Sovereignty Navigation
Voluntary pooling, R2P logic, variable geometry - Outcome: Fair Distribution
Costs and benefits allocated fairly enough for cooperation
Open Questions
- Trade-offs: what if pillars conflict? (e.g., democratic decisions that undermine expertise)
- Enforcement: what if governance lacks power to enforce?
- Exit: how to handle withdrawal from governance frameworks?
- Evolution: how should governance adapt as AI capabilities change?
- Legitimacy thresholds: how much legitimacy is enough?
Conclusion
AI safety governance is not merely a technical challenge. It requires solving problems across six interconnected dimensions:
- Legitimacy: right to govern
- Trust: reliance on actors
- Authority: power to bind
- Democracy: participation in decisions
- Sovereignty: supreme territorial authority
- Distributive justice: fair allocation
Each pillar presents challenges:
- Legitimacy is hard when subjects don't consent
- Trust is fragile when actors have conflicting incentives
- Authority is disputed without clear justification
- Democracy is complicated by technical complexity
- Sovereignty constrains global coordination
- Distributive justice is theoretically contested
The pillars interconnect in complex ways. Legitimacy enables trust and authority. Democracy grounds legitimacy. Sovereignty can conflict with external authority. Distribution affects all other pillars. Effective governance must address all simultaneously.
The unified framework suggests design principles:
- Build all pillars simultaneously
- Design for incompleteness
- Plan for vicious cycles and cultivate virtuous ones
- Use hybrid structures
- Address particularity
- Accept persistent disagreement
The goal is not perfect governance—no such thing exists. The goal is governance that is legitimate enough, trustworthy enough, authoritative enough, democratic enough, sovereignty-compatible enough, and fair enough to prevent catastrophic AI harm while respecting the values that make prevention worthwhile.
Effective AI safety governance requires all six pillars—designed together, for an imperfect world, with no illusions about what's achievable.
References
- Baier, Annette (1986). "Trust and Antitrust."
- Christiano, Thomas (2008). The Constitution of Equality.
- Rawls, John (1993). Political Liberalism.
- Rawls, John (1999). The Law of Peoples.
- Raz, Joseph (1986). The Morality of Freedom.