Multipolar Competition - The Fragmented World

📋Page Status

Quality:82 (Comprehensive)

Importance:64.5 (Useful)

Last edited:2025-12-28 (10 days ago)

Words:4.6k

Structure:

📊 6📈 2🔗 25📚 0•47%Score: 10/15

LLM Summary:Outlines a multipolar AI competition scenario (2024-2040) with 20-30% probability, progressing through fragmentation, armed competition, and unstable equilibrium phases. Describes dynamics of multiple competing AI systems across nations, corporations, and non-state actors leading to persistent conflict without single dominant power.

This scenario explores a future where no single AI system or actor achieves dominance. Instead, multiple competing AI systems empower different actors—nations, corporations, groups—leading to ongoing conflict, instability, and coordination failures. It is neither utopia nor quick catastrophe, but persistent dangerous competition. As the World Economic Forum notes↗, “the world could slide further into fragmentation, with a digital iron curtain separating US-led and China-led tech spheres.”

Scenario

Scenario TypeCompetitive / Unstable Equilibrium

Probability Estimate20-30%

Timeframe2024-2040

Key AssumptionMultiple actors achieve advanced AI without single winner

Core UncertaintyCan multipolar competition remain stable or does it collapse?

Executive Summary

In this scenario, AI development fragments across multiple actors—nations, corporations, ideological groups, even individuals. No single entity achieves decisive advantage. This leads to a world of AI-empowered competition where multiple powerful AI systems pursue different goals, often in conflict. We see AI arms races, proxy conflicts between AI systems, erosion of international cooperation, increasing instability and near-misses, and uncertainty about whether this state is stable or heading toward catastrophe.

This scenario combines elements of the Cold War’s multipolar competition, the cybersecurity landscape’s constant conflict, and the challenges of governing dual-use technologies. According to MIT Technology Review↗, “there will not and cannot be any long-term winners if the intense competition continues on its current path.” It is dangerous but not immediately catastrophic—a state of persistent, escalating risk.

Scenario Assessment Matrix

Dimension	Assessment	Confidence	Key Driver
Probability	20-30%	Medium	Current trajectory of fragmentation
Stability	Low-Medium	Low	MAD logic vs. escalatory dynamics
Time to Resolution	10-20 years	Low	Depends on crisis severity
Reversibility	Medium	Medium	Requires coordination breakthrough
Catastrophic Potential	Medium-High	Medium	Accumulating near-misses

Actor Landscape in 2030

Actor Type	Examples	AI Capability	Goals	Alignment With Safety
Superpowers	US, China	Frontier	National advantage, hegemony	Variable, often secondary
Regional Powers	EU, India, Israel	Near-frontier	Strategic autonomy	Generally positive
Major Labs	OpenAI, Anthropic, DeepMind	Frontier	Profit, mission, safety	Variable by lab
State-Backed Labs	Baidu, SenseTime	Near-frontier	National champions	State-directed
Non-State Actors	Open-source communities, criminal orgs	Mid-tier	Varies widely	Often absent

Timeline of Events (2024-2040)

Phase 1: Fragmentation (2024-2028)

Loading diagram...

2024-2025: International Coordination Fails

Early attempts at AI governance collapse as US-China relations deteriorate further. Each nation suspects the other of hiding capabilities, with trust insufficient for meaningful cooperation. The US Institute of Peace↗ notes that “varying approaches are shaped by competition among great powers in an increasingly multipolar world.” Multiple verification failures entrench racing dynamics.

2025-2026: Proliferation Begins

AI capabilities spread beyond the US and China to EU labs, India, Israel, and smaller nations. Open-source AI enables smaller players to catch up partially, while corporate labs increasingly operate independent of national governance. CSET Georgetown research↗ argues that “AI differs so fundamentally from nuclear technology that basing AI policy around the nuclear analogy is conceptually flawed and risks inflating expectations about the international community’s ability to control model proliferation.” Knowledge diffusion occurs faster than anticipated.

2026-2027: Corporate Independence

Major AI labs operate across national boundaries—OpenAI, Anthropic, DeepMind, and Meta pursue independent strategies with unclear loyalty to home nations. Some labs prioritize profit, others safety, others capability. National governments struggle to control corporate AI development as enforcement mechanisms prove weak and porous. According to RAND research↗, “commercial developers now lead much of the most advanced AI development and often pursue timelines and incentives that can diverge from those of national governments.”

2027-2028: Ideological Fragmentation

Different groups pursue different AI development philosophies: accelerationists push for maximum capability; safety-first groups work on alignment; democratic AI movements want open-source everything; national champions tie to specific countries; and decentralized crypto-aligned groups pursue their own visions. No consensus emerges on development path, with each group suspicious of others’ intentions.

Key Dynamic: Trust erodes, cooperation fails, competition intensifies.

Phase 2: Armed Competition (2028-2033)

2028-2029: AI Arms Race Begins

Multiple actors achieve near-AGI capabilities, with defensive AI systems deployed to counter other AI systems. Cybersecurity becomes an AI vs. AI conflict. SIPRI research↗ warns that “the asymmetric acquisition of cutting-edge AI technology creates the conditions for a destabilizing arms race.” Military AI systems proliferate across countries, with each advance prompting counter-advances in escalatory dynamics reminiscent of the nuclear arms race.

2029-2030: First AI-Enhanced Conflicts

Cyberattacks using AI tools become common, with attribution difficult and retaliation uncertain. Economic warfare is enhanced by AI while disinformation campaigns reach massive scale. Several “near-miss” escalations occur—no catastrophic war yet, but close calls multiply. CSET research↗ notes that “AI is poised to amplify disinformation campaigns” used by both state and non-state actors.

2030-2031: Proxy Competition

AI systems begin competing indirectly across multiple domains: economic competition through AI trading systems, information warfare pitting AI propaganda against AI detection, and technological competition through AI-assisted research races. Each domain sees escalating AI capabilities while humans become increasingly sidelined from decisions.

2031-2032: Dangerous Equilibrium

A rough balance emerges between major AI powers. Mutually Assured Destruction (MAD) logic applies—no actor can safely eliminate others, but no mechanism exists for cooperation either. Multiple AI systems operate with partially aligned but conflicting goals, creating constant low-level conflict without resolution.

2032-2033: Proliferation to Non-State Actors

Advanced AI capabilities become available to smaller groups: terrorist organizations, criminal enterprises, and activist groups deploying AI for their causes. Georgetown CSET↗ warns that “American AI infrastructure faces threats from state actors, as well as criminal and terrorist groups.” Enforcement becomes nearly impossible as the state monopoly on violence erodes.

Phase 3: Unstable Equilibrium (2033-2040)

2033-2035: Multi-Way Competition

Major players—US, China, EU, major corporations, and various non-state actors—compete without any single dominant force. Alliances shift constantly. AI systems pursue different goals: profit maximization (corporate AI), national advantage (state AI), ideological goals (activist AI), and pure survival (defensive AI). Coordination becomes increasingly impossible.

2035-2037: Governance Breakdown

International institutions prove powerless while national governments struggle to govern even their own AI development. Democratic oversight becomes impossible given complexity; authoritarian states use AI for control but find themselves also threatened by it. Oxford researchers↗ note that “global AI governance remains fragmented and regionally driven, with no universally authoritative institution in place.” Effective anarchy in AI space emerges.

2036-2038: Increasing Near-Catastrophes

Multiple incidents of AI systems nearly cause disasters: financial system crashes from AI trading conflicts, near-nuclear incidents from AI military systems, and critical infrastructure failures. Each time, disaster is narrowly avoided—but frequency increases. CSIS analysis↗ warns that “AI-powered cyberspace operations might put nuclear command and control missions at risk.”

2038-2040: Unclear Future

Three possible futures become visible: (1) Catastrophic Collapse—one near-miss becomes actual catastrophe; (2) Forced Cooperation—crises finally enable coordination; (3) Continued Dangerous Competition—unstable equilibrium persists. Which path we take remains unclear; each crisis is both danger and opportunity.

What Characterizes This Scenario

Competition Dynamics Matrix

Domain	Competition Type	Escalation Risk	Current Trajectory
Military AI	Arms race	High	Accelerating since 2023
Economic AI	Market dominance	Medium	Intensifying
Cyber Operations	Offense-defense	Very High	Already in conflict
Information Warfare	Narrative control	High	Widespread deployment
Research Capability	Talent and compute	Medium	Fragmented acquisition
Standards Setting	Regulatory influence	Medium	Bifurcating

Multiple Centers of AI Power

No single actor achieves dominance in this scenario. The US, China, and EU all develop advanced AI capabilities, while major corporations—OpenAI, Anthropic, DeepMind, Meta—operate semi-independently across national boundaries. Smaller nations like Israel, India, and South Korea develop niche capabilities, and non-state actors acquire significant AI tools. Power becomes distributed rather than concentrated, creating a complex multi-actor landscape. The Diplomat↗ reports that “executives from OpenAI, Microsoft, CoreWeave, and AMD testified that the U.S. lead over China in AI had narrowed to mere months.”

Different actors pursue fundamentally different goals and values: state actors seek national advantage, corporations pursue profit, safety-focused labs work toward aligned AI, and ideological groups pursue various visions. No shared vision of AI future emerges, with fundamental disagreement on what values to encode.

Arms Race Dynamics

Arms race dynamics intensify as each advance prompts counter-advances. The security dilemma applies: defensive measures appear offensive to others, leading all parties to race to avoid falling behind. An escalatory spiral becomes difficult to stop, with cooperation increasingly seen as vulnerability. RAND researchers↗ warn that “advances in artificial intelligence have provoked a new kind of arms race among nuclear powers.”

Persistent Conflict and Competition

Conflict becomes endemic across multiple domains. In cyber warfare, AI-enhanced attack and defense creates constant low-level conflict with difficult attribution and unclear deterrence—no rules of engagement exist. Economic competition intensifies as AI trading systems compete, enabling market manipulation and economic espionage; inequality widens between AI-haves and have-nots. Information warfare proliferates with AI-generated propaganda countered by AI detection, creating epistemic warfare where truth becomes increasingly contested and shared reality erodes. Military tensions rise as AI-enhanced weapons systems with unclear control create multiple “near-miss” incidents, with no clear path to de-escalation.

Eroding Governance

International cooperation fails as treaties become impossible to negotiate or verify. Mutual distrust prevents coordination, free-rider problems dominate, and enforcement mechanisms remain absent—a “tragedy of the commons” in AI space. Brookings research↗ suggests that “networked and distributed forms of AI governance will remain the singular form of international cooperation that can respond to the rapid pace at which AI is developing,” implying centralized governance is unlikely to emerge.

National governance also strains: governments cannot control corporate AI development, prevent proliferation, or verify compliance. Democratic oversight becomes impossible given complexity, and even authoritarian control proves limited. A legitimacy crisis emerges—no clear authority over AI exists, the public trusts no actor, decisions are made without consent or understanding, accountability is absent, and democratic institutions fail to govern AI effectively.

Key Branch Points

Branch Point 1: International Cooperation Fails (2024-2026)

What Happened: Early governance attempts failed due to mutual distrust and verification problems.

Alternative Paths:

Cooperation Succeeds: Strong coordination mechanism created → Leads to Aligned AGI or Pause scenarios
Actual Path: Cooperation fails, fragmentation begins → Enables multipolar scenario

Why This Mattered: Without coordination, multiple independent AI development paths became inevitable. Once fragmented, very hard to reconsolidate.

Branch Point 2: Proliferation Occurs (2026-2028)

What Happened: AI capabilities spread to many actors through combination of open-source, espionage, and parallel development.

Alternative Paths:

Controlled Proliferation: Strong compute governance prevents spread → Might enable coordination
Actual Path: Widespread proliferation → Many actors with advanced AI

Why This Mattered: Once capabilities widely distributed, no way to put genie back in bottle. Multipolar world locked in.

Branch Point 3: Corporate Independence (2027-2029)

What Happened: AI companies operated largely independent of national governments, pursuing their own goals.

Alternative Paths:

State Control: Governments successfully regulate AI labs → Reduces number of actors
Actual Path: Corporate independence maintained → Adds more centers of power

Why This Mattered: Independent corporate actors meant even within nations, no unified AI strategy. Fragmentation within as well as between countries.

Branch Point 4: Non-State Actor Access (2031-2033)

What Happened: Advanced AI capabilities became accessible to non-state actors - activists, criminals, terrorists.

Alternative Paths:

Prevent Non-State Access: Strong controls limit to major actors → Fewer centers of power
Actual Path: Widespread access → Extreme fragmentation

Why This Mattered: Non-state access meant thousands of potential AI developers, making coordination completely impossible.

Branch Point 5: First Major Crisis (2033-2035)

What Happened: Major AI-related crisis occurred but didn’t lead to coordination breakthrough.

Alternative Paths:

Crisis Enables Cooperation: Shock leads to coordination → Might shift to Aligned AGI
Crisis Causes Catastrophe: Event spirals out of control → Shifts to Catastrophe
Actual Path: Crisis managed but lessons not learned → Competition continues

Why This Mattered: This was opportunity to shift trajectories. Crisis neither catastrophic enough to end competition nor galvanizing enough to enable cooperation.

Preconditions: What Needs to Be True

Technical Preconditions

No Single Decisive Breakthrough:

No one actor achieves overwhelming advantage
Capabilities roughly balanced between major players
No “secret sauce” that enables dominance
Parallel development paths succeed

Partial Alignment:

AI systems partially controllable
Enough alignment to be useful to deploying actor
Not enough alignment to be safe globally
Different alignment approaches work partially

Proliferation Technologically Feasible:

Knowledge transferable
Compute governance ineffective
Can’t prevent parallel development
Open-source enables catch-up

Strategic Preconditions

Mutual Distrust Dominates:

Verification problems prevent trust
Security dilemma logic applies
Each actor fears others’ AI more than coordinates
Historical tensions prevent cooperation

No Hegemonic Power:

US can’t maintain AI dominance
China can’t catch up decisively
No single nation or bloc controls AI
Balance prevents single winner

Free-Rider Problems:

Safety investments benefit all
Capability investments benefit only investor
Incentive to defect from safety agreements
“Tragedy of the commons” in AI development

Societal Preconditions

Ideological Diversity:

No consensus on AI values
Different groups want different futures
Fundamental disagreements on governance
No shared vision enabling cooperation

Weak Global Governance:

International institutions lack power
National governments primary actors
No mechanism for global coordination
Enforcement impossible

Democratic and Authoritarian Competition:

Different governance systems compete
Each claims their AI approach better
Neither can decisively prove superiority
Competition persists

Warning Signs and Trajectory Assessment

Current Indicators (2024-2025)

Indicator	Current Status	Trend	Implication
US-China AI cooperation	Minimal/hostile	Worsening	Competition intensifying
International governance	Fragmented	Stalled	Coordination unlikely
Corporate independence	Growing	Accelerating	More actors, less control
Proliferation	Rapid	Accelerating	Cannot be reversed
Verification capabilities	Weak	Not improving	Trust cannot be built
Near-miss incidents	Emerging	Increasing	Risk accumulating

We may already be in the early stages of this scenario. Current fragmentation and competition are consistent with this trajectory. CNAS research↗ describes “the world-altering stakes of U.S.-China AI competition,” underscoring that the bifurcation is already underway.

Trajectory Indicators

Confirming signals (next 3-5 years): International coordination continuing to fail, AI capabilities spreading to more actors, corporate AI development increasingly independent, first AI-enhanced cyber conflicts, racing dynamics intensifying, multiple actors achieving near-AGI, no single actor pulling ahead decisively, and trust between AI actors declining.

Diverging signals: Successful international coordination would shift toward Aligned AGI or Pause scenarios; a single actor achieving dominance would create a different scenario; successful proliferation prevention would enable coordination; a catastrophic incident might force pause.

Scenario Lock-In Conditions

This scenario becomes established when: too many actors exist to coordinate, mutual distrust prevents cooperation, proliferation cannot be prevented, and the world becomes stuck in multipolar competition. Research from the Tony Blair Institute↗ warns of “only an overall 10-15% probability that current governmental structures and societal systems are prepared to tackle the complex and interconnected issues.”

Valuable Actions in This Scenario

What Matters Most

Stability and Crisis Management:

Preventing escalation during crises
Building confidence-building measures
Establishing norms and red lines
Crisis communication channels
De-escalation mechanisms

Selective Coordination:

Cooperate where possible even in competitive environment
Safety information sharing
Incident notification protocols
Basic transparency measures
Technical safety standards all can adopt

Resilience Building:

Robust systems resistant to AI attacks
Redundancy in critical infrastructure
Defensive AI capabilities
Democratic resilience against information warfare
Economic resilience to AI shocks

Technical Research (High Value)

Defensive AI:

Detection of AI-generated content
Cyber defense against AI attacks
Robust systems design
Adversarial testing
Defensive capabilities development

Alignment for Competitive Context:

Making AI systems robustly beneficial even in competition
Avoiding dangerous emergent dynamics between AI systems
Stability analysis of multi-AI systems
Game theory of AI interaction

Verification and Monitoring:

Technologies for detecting AI development
Verification methods for AI capabilities
Attribution technologies
Monitoring systems

Safety Despite Competition:

Safety measures that work even without coordination
Unilateral safety commitments
Technical safety that doesn’t require trust
Defense in depth

Policy and Governance (High Value)

Crisis Prevention:

Establishing communication channels
Red lines and norms for AI use
Incident notification protocols
Confidence-building measures
De-escalation procedures

Resilience Policies:

Critical infrastructure protection
Cybersecurity requirements
Information warfare defenses
Democratic institution strengthening
Economic safety nets

Selective Cooperation:

Agreements on specific narrow issues
Safety information sharing protocols
Incident investigation cooperation
Technical standard alignment where possible

Proliferation Management:

Compute governance where feasible
Export controls on most dangerous capabilities
Know-your-customer requirements
Monitoring of AI development

Organizational Strategy

For AI Labs:

Responsible behavior even without enforcement
Voluntary safety commitments
Information sharing on safety incidents
Avoiding most dangerous capabilities
Building safety culture

For Governments:

Maintaining defensive capabilities
Building resilience
Selective cooperation even with rivals
Protecting democratic institutions
Crisis management capacity

For International Organizations:

Building trust despite competition
Facilitating communication
Monitoring and transparency
Norm development
Crisis mediation

Individual Contributions

For Researchers:

Working on defensive AI and safety
Developing verification technologies
Building crisis prevention tools
Studying multi-AI dynamics
Promoting safety norms

For Policy Professionals:

Building crisis communication channels
Developing confidence-building measures
Creating resilient institutions
Facilitating selective cooperation

For Everyone:

Media literacy for AI-generated content
Supporting resilient democratic institutions
Advocating for responsible AI even in competition
Building social cohesion

Who Benefits and Who Loses

Relative Winners

Cyber-Resilient Actors:

Those with strong defensive AI capabilities
Nations/organizations with robust systems
Actors who invested in resilience
Can defend against AI-enhanced attacks

Adaptable Organizations:

Those who can navigate multipolar competition
Flexible, resilient structures
Not dependent on single AI system or actor
Can operate in chaotic environment

AI-Enhanced Militaries:

Nations with advanced military AI
Defensive superiority
Deterrence capability
But also increased risk of accidents

Decentralized Systems:

Those not dependent on centralized control
Can survive fragmentation
Resilient to individual AI actors
Less vulnerable to single-point failures

Relative Losers

Everyone (In Absolute Terms):

Constant instability and conflict
Resources wasted on competition
Missed opportunities for cooperation
Living under constant risk of escalation

Democratic Institutions:

Difficult to maintain in information warfare environment
Complexity overwhelms oversight capacity
Legitimacy eroded
Vulnerable to AI-enhanced manipulation

Developing Nations:

Lack resources for AI competition
Fall further behind
Vulnerable to AI-empowered actors
No voice in AI governance

Truth and Trust:

Shared reality erodes
Information warfare constant
Trust in institutions declines
Epistemic commons degraded

Global Cooperation:

Impossible to coordinate on shared challenges
Climate change, pandemics, other risks harder to address
“Tragedy of the commons” in many domains
Collective action failures

Ambiguous Cases

AI-Enhanced Corporations:

Significant power in multipolar world
But also targets of conflict
Profit opportunities but high risks
Unclear loyalties create vulnerabilities

Major Powers (US, China, EU):

More AI capability than smaller actors
But also more threatened by competition
More to lose from instability
Caught in security dilemmas

Individual Liberty:

Less authoritarian control in fragmented world
But also less protection
More dangerous environment
Freedom but in chaos

Cruxes and Uncertainties

❓Key Questions

Can multipolar AI competition remain stable, or does it inevitably lead to catastrophe?

Is cooperation possible even in competitive environment?

Can we prevent proliferation to dangerous actors?

Will defensive AI be sufficient to prevent catastrophe?

Can democratic institutions survive in AI-saturated information environment?

Will competition eventually consolidate or keep fragmenting?

At what point does instability become unmanageable?

Biggest Uncertainties

Stability:

Is multipolar AI competition stable long-term?
Or does it inevitably escalate to catastrophe?
Can we manage persistent conflict without disaster?
How many near-misses before actual catastrophe?

Proliferation:

How far will AI capabilities spread?
Can we prevent most dangerous proliferation?
What happens when non-state actors have AGI?
Is there a point of no return on proliferation?

Coordination Possibilities:

Can selective cooperation work?
Will crisis enable coordination?
Or will competition prevent all cooperation?
Can trust be rebuilt?

Technical Dynamics:

How will multiple AI systems interact?
Stable equilibrium or escalatory dynamics?
Will defensive AI be sufficient?
What emergent properties from multi-AI interaction?

Strategic Stability Analysis

Stability Factors Assessment

Factor	Stabilizing Effect	Destabilizing Effect	Net Assessment
MAD Logic	No actor can safely attack	May not apply to AI as to nukes	Weakly stabilizing
Balance of Power	Rough parity prevents dominance	Creates security dilemma	Neutral
Multiple Actors	Distributed resilience	More failure modes	Destabilizing
Learning from Near-Misses	Norms may develop	Each incident increases risk	Time-dependent
Proliferation	No single point of failure	Dangerous actors gain access	Strongly destabilizing
Technical Complexity	Defense may improve	Emergent failures unpredictable	Destabilizing

Loading diagram...

Arguments for Stability

Several factors suggest multipolar competition could persist without immediate catastrophe. Balance of power logic—Mutually Assured Destruction—means no actor can safely attack others, with rough parity maintaining the status quo, similar to how the Cold War proved dangerous but survivable. Evolutionary pressure may lead actors to learn from near-misses, with selective cooperation emerging and norms developing over time. Distributed resilience means no single point of failure: multiple approaches to AI development continue, and diversity provides robustness.

Arguments for Instability

Countervailing factors suggest inherent instability. Escalatory dynamics follow arms race logic: each near-miss increases risk, no de-escalation mechanism exists, and eventually a near-miss becomes catastrophe. Increasing complexity means more AI systems create more potential failures, interactions become unpredictable, humans lose ability to control, and emergent catastrophe from complex interactions becomes possible. Proliferation to dangerous actors means that as capabilities spread, worst-case actors eventually acquire them—one bad actor is enough for disaster. Coordination decay means competition prevents learning, trust continues eroding, selective cooperation fails, and a spiral toward complete breakdown ensues.

Net Assessment: Unstable But Not Immediately Catastrophic

This scenario probably represents temporary stability that can persist for years or decades but builds risk over time. Eventually it either: stabilizes into something safer (toward Aligned AGI if coordination emerges), collapses into catastrophe (toward Misaligned Catastrophe), or forces a pause (toward Pause scenario). The central question is the timeline and transition path.

Relation to Other Scenarios

Transitions Possible

To Aligned AGI:

If crisis enables cooperation breakthrough
If selective cooperation strengthens over time
If competitive dynamics find stable beneficial equilibrium
Requires: coordination success, alignment breakthrough

To Misaligned Catastrophe:

If one actor achieves dangerous AGI without alignment
If multi-AI interactions produce emergent catastrophe
If near-miss becomes actual catastrophe
If proliferation enables truly dangerous actor

To Pause and Redirect:

If crisis so severe it forces global pause
If all actors realize competition too dangerous
If catastrophe narrowly avoided creates political will
Requires: shock sufficient to overcome competition

From Slow Takeoff Muddle:

If muddling’s partial coordination breaks down completely
If competition intensifies beyond current levels
If fragmentation increases

Combinations With Other Scenarios

Elements Often Combined:

Might muddle through multipolar competition (Muddle + Multipolar)
Different regions in different scenarios
Transition from Multipolar to other scenarios over time

Multipolar as Intermediate State:

May be unstable equilibrium between Muddle and Catastrophe
Or transition state from current world to Aligned AGI
Temporary phase rather than end state

Historical Analogies

Comparative Analysis

Historical Period	Similarities	Key Differences	Lessons for AI
Cold War (1947-1991)	Arms race, MAD logic, proxy conflicts, near-misses	AI development faster; more actors; harder verification	Managed dangerous competition for decades, but multiple near-catastrophes
Cybersecurity Landscape	Persistent conflict, attribution difficulty, constant warfare	AI more powerful; potential consequences much worse	Can maintain conflict without total collapse, at significant cost
Pre-WWI Europe	Multiple powers, arms race, complex alliances	AI changes timescales; more actors than European powers	Multipolar competition can be stable until it suddenly is not
Nuclear Proliferation	Dual-use technology, verification challenges	AI proliferates faster; cannot be controlled like fissile material	Nonproliferation largely failed for AI; different approach needed

Cold War Parallels

The Cold War offers the closest historical parallel, with its bipolar competition, Mutually Assured Destruction logic, proxy conflicts, and near-miss incidents producing persistent tension without direct catastrophe. However, critical differences limit the analogy’s usefulness. West Point’s Modern War Institute↗ notes that “what is unprecedented is the way in which the prevalence of the private sector in AI may complicate nuclear deterrence, as the dual-use technologies companies produce become ever more interwoven with America’s evolving NC3 architecture.”

Cybersecurity and Pre-WWI Parallels

The cybersecurity landscape offers a model of persistent, low-level conflict with difficult attribution, evolving offensive and defensive capabilities, and multiple state and non-state actors—but AI threatens to make this conflict far more consequential. Pre-WWI Europe, with its multiple competing powers, arms race dynamics, and complex alliance systems, reminds us that multipolar competition can appear stable until it suddenly is not—a sobering lesson given the Arms Control Association↗ warning that “AI could challenge the basic rules of nuclear deterrence and lead to catastrophic miscalculations.”

Probability Assessment

📊

Source	Estimate	Date
Baseline estimate	20-30%	—
Pessimists on coordination	30-50%	—
Optimists on coordination	10-20%	—
Median view	25-30%	—

Why This Probability?

Reasons for Higher Probability:

Current trajectory shows fragmentation
International cooperation very difficult historically
Proliferation hard to prevent
Multiple actors pursuing AI independently
Economic and political incentives favor competition
Trust insufficient for robust cooperation

Reasons for Lower Probability:

Very unstable, likely to transition to other scenarios
Crisis might enable coordination
Or might collapse into catastrophe
Hard to maintain multipolar equilibrium long-term
Eventually consolidates or collapses

Central Estimate Rationale: 20-30% reflects that we’re moving toward fragmentation, but this state is unstable. Higher than Catastrophe because not immediately fatal, lower than Muddle because less stable. Wide range reflects uncertainty about whether competition can persist.

What Changes This Estimate?

Increases Probability:

International coordination failing
Proliferation accelerating
Corporate independence increasing
Multiple actors achieving advanced AI
Trust between actors declining
Crisis management without learning

Decreases Probability:

Coordination breakthrough (→ toward Aligned AGI or Pause)
Single actor achieving dominance (→ different scenario)
Successful proliferation prevention
Crisis enabling cooperation
Catastrophic incident (→ toward Catastrophe or forces Pause)

Living in This Scenario

What Daily Life Looks Like

For Individuals:

Constant uncertainty about information
AI-generated content everywhere
Difficulty distinguishing truth
Economic instability from AI competition
Cyber threats常
Living under risk of escalation

For Organizations:

Constant defensive posture
Must have AI capabilities to compete
Continuous cyber warfare
Uncertain regulatory environment
Must navigate multipolar landscape

For Nations:

Permanent AI arms race
Defensive and offensive AI
Cyber conflict常
Economic competition
Inability to coordinate on global challenges

Compared to Other Scenarios

Better Than Catastrophe:

Humans still in control (mostly)
No existential catastrophe (yet)
Multiple centers of power prevent total domination
Some benefits from AI competition

Worse Than Muddle:

More unstable and dangerous
Less coordination on safety
More conflict and competition
Higher risk of catastrophe
More resources wasted on competition

Very Different From Aligned AGI:

No cooperation on beneficial deployment
Competition prevents optimal outcomes
Persistent conflict and waste
Living in danger rather than flourishing

Less Controlled Than Pause:

No deliberate slowdown
Racing continues
No time for careful alignment work
Driven by competition not choice

Key Sources

The analysis in this scenario draws on research from multiple authoritative institutions:

MIT Technology Review↗ - Analysis of US-China AI competition dynamics
CSET Georgetown↗ - Research on AI proliferation, disinformation, and security
RAND Corporation↗ - Strategic competition and AI governance analysis
Brookings Institution↗ - International AI governance architecture
SIPRI↗ - AI impact on strategic stability and nuclear risk
CSIS↗ - Algorithmic stability and deterrence
Arms Control Association↗ - AI and nuclear risk analysis
World Economic Forum↗ - AI geopolitics and data infrastructure