Skip to content
Longterm Wiki
Navigation
Updated 2025-12-28HistoryData
Page StatusContent
Edited 3 months ago2.2k words5 backlinksUpdated quarterlyOverdue by 8 days
64QualityGood •47ImportanceReference67.5ResearchModerate
Content9/13
SummaryScheduleEntityEdit historyOverview
Tables24/ ~9Diagrams3/ ~1Int. links39/ ~18Ext. links14/ ~11Footnotes0/ ~7References19/ ~7Quotes0Accuracy0RatingsN:4.5 R:6 A:7.5 C:7.5Backlinks5
Issues3
QualityRated 64 but structure suggests 100 (underrated by 36 points)
Links12 links could use <R> components
StaleLast edited 98 days ago - may need review
TODOs4
Complete 'Conceptual Framework' section
Complete 'Quantitative Analysis' section (8 placeholders)
Complete 'Strategic Importance' section
Complete 'Limitations' section (6 placeholders)

AI Risk Portfolio Analysis

Analysis

AI Risk Portfolio Analysis

Quantitative portfolio framework recommending AI safety resource allocation: 40-70% to misalignment, 15-35% to misuse, 10-25% to structural risks, varying by timeline. Based on 2024 funding analysis ($110-130M total), identifies specific gaps including governance (underfunded by $15-20M), agent safety ($7-12M gap), and international capacity ($11-16M gap).

Model TypePrioritization Framework
FocusResource Allocation
Key OutputRisk magnitude comparisons and allocation recommendations
Related
Analyses
AI Compounding Risks Analysis ModelFlash Dynamics Threshold Model
2.2k words · 5 backlinks

Overview

This framework provides quantitative estimates for allocating limited resources across AI risk categories. Based on expert surveys and risk assessment methodologies from organizations like RAND and Center for Security and Emerging Technology (CSET), the analysis estimates misalignment accounts for 40-70% of existential risk, misuse 15-35%, and structural risks 10-25%.

The model draws from portfolio optimization theory and Coefficient Giving's cause prioritization framework, addressing the critical question: How should the AI safety community allocate its $100M+ annual resources across different risk categories? All estimates carry substantial uncertainty (±50% or higher), making the framework's value in relative comparisons rather than precise numbers.

Risk Assessment Matrix

Risk CategoryX-Risk ShareP(Catastrophe)TractabilityNeglectednessCurrent Allocation
Misalignment40-70%15-45%2.5/53/5≈50%
Misuse15-35%8-25%3.5/54/5≈25%
Structural10-25%5-15%4/54.5/5≈15%
Accidents (non-X)5-15%20-40%4.5/52.5/5≈10%
Uncertainty Bounds

These estimates represent informed speculation based on limited data. Superforecasters and AI experts show significant disagreement on these parameters, with confidence intervals often spanning 2-3x.

Strategic Prioritization Framework

Expected Value Calculation

The framework applies standard expected value methodology:

Priority Score=Risk Magnitude×P(Success)×Neglectedness Multiplier\text{Priority Score} = \text{Risk Magnitude} \times \text{P(Success)} \times \text{Neglectedness Multiplier}
CategoryRisk MagnitudeP(Success)NeglectednessPriority Score
Misalignment8.5/100.250.61.28
Misuse6.0/100.350.81.68
Structural4.5/100.400.91.62

Timeline-Dependent Allocation

Resource allocation should vary significantly based on AGI timeline beliefs:

Timeline ScenarioMisalignmentMisuseStructuralRationale
Short (2-5 years)70-80%15-20%5-10%Only time for direct alignment work
Medium (5-15 years)50-60%25-30%15-20%Balanced portfolio approach
Long (15+ years)40-50%20-25%25-30%Time for institutional solutions
Diagram (loading…)
pie title Current Resource Allocation vs Optimal (Medium Timeline)
  "Misalignment (Current 50%)" : 50
  "Misalignment (Optimal 55%)" : 55
  "Misuse (Current 25%)" : 25
  "Misuse (Optimal 27%)" : 27
  "Structural (Current 15%)" : 15
  "Structural (Optimal 18%)" : 18

Marginal Value Analysis

Current Bottlenecks by Risk Category

CategoryPrimary BottleneckMarginal $ ValueSaturation RiskKey Organizations
MisalignmentConceptual clarityHigh (if skilled)MediumMIRI, Anthropic
MisuseGovernment engagementVery HighLowCNAS, CSET
StructuralFramework developmentHighVery LowGovAI, CAIS
AccidentsImplementation gapsMediumHighPartnership on AI

Funding Landscape Analysis

Based on comprehensive analysis from Coefficient Giving, Longview Philanthropy estimates, and LTFF reporting, external AI safety funding reached approximately $110-130M in 2024:

Funding Source2024 AmountShareKey Focus Areas
Coefficient Giving$63.6M≈49%Technical alignment, evaluations, governance
Survival & Flourishing Fund$19M+≈15%Diverse safety research
Long-Term Future Fund$5.4M≈4%Early-career, small orgs
Jaan Tallinn & individual donors$20M≈15%Direct grants to researchers
Government (US/UK/EU)$32.4M≈25%Policy-aligned research
Other (foundations, corporate)$10-20M≈10%Various

The breakdown by research area reveals significant concentration in interpretability and evaluations:

Research Area2024 FundingShareTrendOptimal (Medium Timeline)
Interpretability$52M40%Growing30-35%
Evaluations/benchmarking$23M18%Rapid growth15-20%
Constitutional AI/RLHF$38M29%Stable25-30%
Governance/policy$18M14%Underfunded20-25%
Red-teaming$15M12%Growing10-15%
Agent safety$8.2M6%Emerging10-15%
Coefficient Giving Dominance

Coefficient Giving accounts for nearly 60% of all external AI safety investment, with $63.6M deployed in 2024. Since 2017, Coefficient Giving has donated approximately $336M to AI safety (~12% of their total $2.8B in giving). The median Coefficient Giving AI safety grant is $257k; the average is $1.67M.

Risk Interdependency Network

Rather than independent categories, risks exhibit complex interactions affecting prioritization:

Diagram (loading…)
flowchart TD
  CAP[AI Capabilities] -->|amplifies| MIS[Misalignment Risk]
  CAP -->|enables| USE[Misuse Risk]
  USE -->|degrades| GOV[Governance Quality]
  GOV -->|mitigates| USE
  GOV -->|weakly mitigates| MIS
  STR[Structural Risks] -->|erodes| GOV
  MIS -->|if realized| STR
  
  style CAP fill:#ff9999
  style MIS fill:#ffcccc
  style USE fill:#ffffcc
  style GOV fill:#ccffcc
  style STR fill:#ccccff

Correlation Matrix

Risk PairCorrelationImplication for Portfolio
Misalignment ↔ Capabilities+0.8High correlation; capabilities research affects risk
Misuse ↔ Governance Quality-0.6Good governance significantly reduces misuse
Structural ↔ All Others+0.4Structural risks amplify other categories

Comparative Assessment Methods

Expert Survey Results

Multiple surveys reveal substantial disagreement on AI risk magnitude. AI Impacts 2022 expert survey of 738 AI researchers and the Conjecture internal survey provide contrasting perspectives:

Risk CategoryAI Impacts MedianConjecture MedianExpert Disagreement (IQR)Notes
Total AI X-risk5-10%80%2-90%Massive disagreement
Misalignment-specific25%60%+10-50%Safety org workers higher
Misuse (Bio/weapons)15%30-40%5-35%Growing concern
Economic Disruption35%50%+20-60%Most consensus
Authoritarian Control20%40%8-45%Underexplored
Interpreting Survey Disagreement

The Conjecture survey (N=22 AI safety researchers) found no respondent reported less than 10% extinction risk, with a median of 80%. However, this sample has severe selection bias—AI safety researchers self-select for high risk estimates. The AI Impacts survey sampled mainstream ML researchers with lower risk estimates but suffered from non-response bias. True uncertainty likely spans 2-50% for catastrophic outcomes.

Case Study Comparisons

Historical technology risk portfolios provide calibration:

TechnologyPrimary Risk FocusSecondary RisksOutcome Assessment
Nuclear weaponsAccident prevention (60%)Proliferation (40%)Reasonable allocation
Climate changeMitigation (70%)Adaptation (30%)Under-weighted adaptation
Internet securityTechnical fixes (80%)Governance (20%)Under-weighted governance

Pattern: Technical communities systematically under-weight governance and structural interventions.

Uncertainty Analysis

Key Cruxes Affecting Allocation

Key Questions

  • ?What's the probability of transformative AI by 2030? (affects all allocations)
  • ?How tractable is technical alignment with current approaches?
  • ?Does AI lower bioweapons barriers by 10x or 1000x?
  • ?Are structural risks primarily instrumental or terminal concerns?
  • ?What's the correlation between AI capability and alignment difficulty?

Sensitivity Analysis

Parameter ChangeEffect on Misalignment PriorityEffect on Misuse Priority
Timeline -50% (shorter)+15-20 percentage points-5-10 percentage points
Alignment tractability +50%-10-15 percentage points+5-8 percentage points
Bioweapons risk +100%-5-8 percentage points+10-15 percentage points
Governance effectiveness +50%-3-5 percentage points+8-12 percentage points

Geographic Distribution of Funding

The AI safety funding landscape shows significant geographic concentration, with implications for portfolio diversification:

Region2024 FundingShareKey OrganizationsGap Assessment
SF Bay Area$48M37%CHAI, MIRI, AnthropicWell-funded
London/Oxford$32M25%FHI, DeepMind, GovAIWell-funded
Boston/Cambridge$12M9%MIT, HarvardGrowing
Washington DC$8M6%CSET, CNAS, BrookingsPolicy focus
Rest of US$10M8%Academic dispersedModerate
Europe (non-UK)$8M6%Berlin, Zurich hubsUnderfunded
Asia-Pacific$4M3%Singapore, AustraliaSeverely underfunded
Rest of World$8M6%VariousVery limited
Emerging Hubs

Government initiatives are expanding geographic coverage: Canada's $12M AI Safety Research Initiative, Australia's $8.4M Responsible AI Program, and Singapore's $5.6M AI Ethics Research Fund launched in 2024-2025. These represent opportunities for funding diversification beyond the US/UK axis.

Diagram (loading…)
flowchart LR
  subgraph US["United States (54%)"]
      SF[SF Bay Area<br/>37%]
      DC[DC<br/>6%]
      BOS[Boston<br/>9%]
  end
  subgraph UK["United Kingdom (25%)"]
      LON[London/Oxford<br/>25%]
  end
  subgraph Other["Other (21%)"]
      EU[Europe<br/>6%]
      APAC[Asia-Pacific<br/>3%]
      ROW[Rest of World<br/>12%]
  end

  style SF fill:#90EE90
  style LON fill:#90EE90
  style EU fill:#FFB6C1
  style APAC fill:#FF6B6B
  style ROW fill:#FFB6C1

Implementation Recommendations

For Major Funders

Based on 2024 funding analysis, specific portfolio rebalancing recommendations:

Funder TypeCurrent AllocationRecommended ShiftSpecific OpportunitiesPriority
Coefficient Giving68% evals, 12% interp+15% governance, +10% agent safetyGovAI expansion, international capacityHigh
SFF/individual donorsTechnical focus+$5-10M to neglected areasValue learning, formal verificationHigh
LTFFEarly career, small orgsMaintain current portfolioContinue diversified approachMedium
Government agenciesPolicy-aligned research+$20-30M to independent oversightAISI expansion, red-teamingVery High
Tech philanthropistsVaries widelyCoordinate via giving circlesReduce duplicationMedium

Specific Funding Gaps (2025):

Gap AreaCurrent FundingOptimalGapRecommended Recipients
Agent safety$8.2M$15-20M$7-12MMETR, Apollo, academic groups
Value alignment theory$6.5M$12-15M$5-9MMIRI, academic philosophy
International capacity$4M$15-20M$11-16MNon-US/UK hubs
Governance research$18M$25-35M$7-17MGovAI, CSET, Brookings
Red-teaming$15M$20-25M$5-10MIndependent evaluators

For Research Organizations

Capability-Building Priorities:

Organization SizePrimary FocusSecondary FocusRationale
Large (>50 people)Maintain current specializationAdd governance capacityComparative advantage
Medium (10-50 people)70% core competency30% neglected areasDiversification benefits
Small (<10 people)Focus on highest neglectednessNoneResource constraints

For Individual Researchers

Career decision framework based on 80,000 Hours methodology:

Career StageIf Technical BackgroundIf Policy BackgroundIf Economics/Social Science
Early (0-5 years)Alignment researchMisuse preventionStructural risk analysis
Mid (5-15 years)Stay in alignment vs. pivotGovernment engagementInstitution design
Senior (15+ years)Research leadershipPolicy implementationField coordination

Current State and Trajectory

2024 Funding Landscape

Based on detailed analysis and Coefficient Giving grant data, external AI safety funding has evolved significantly:

YearExternal FundingInternal Lab SafetyTotal (Est.)Key Developments
2020$40-60M$50-100M$100-160MCoefficient Giving ramping up
2021$60-80M$100-200M$160-280MAnthropic founded
2022$80-100M$200-400M$280-500MChatGPT launch
2023$90-120M$400-600M$490-720MMajor lab investment
2024$110-130M$500-700M$610-830MGovernment entry
Internal vs External Funding

Major AI labs—Anthropic, OpenAI, and DeepMind—invest an estimated $500M+ combined in internal safety research annually, dwarfing external philanthropic funding. However, internal research may face conflicts of interest with commercial objectives, making external independent funding particularly valuable for governance and red-teaming work.

Coefficient Giving Technical AI Safety Grants (2024)

Detailed analysis of Coefficient Giving's $28M in Technical AI Safety grants reveals:

Focus AreaShare of CG TAISKey RecipientsAssessment
Evaluations/benchmarking68%METR, Apollo, UK AISIHeavily funded
Interpretability12%Anthropic, RedwoodWell-funded
Robustness8%Academic groupsModerate
Value alignment5%MIRI, academicUnderfunded
Field building5%MATS, training programsAdequate
Other approaches2%VariousExploratory

Projected 2025-2027 Needs

ScenarioAnnual NeedTechnicalGovernanceField BuildingRationale
Short timelines (2-5y)$300-500M70%20%10%Maximize alignment progress
Medium timelines (5-15y)$200-350M55%30%15%Build institutions + research
Long timelines (15+y)$150-250M45%35%20%Institutional capacity

Coefficient Giving's 2025 RFP commits at least $40M to technical AI safety, with potential for "substantially more depending on application quality." Priority areas marked include agent safety, interpretability, and evaluation methods.

Key Model Limitations

What This Framework Doesn't Capture

LimitationImpact on RecommendationsMitigation Strategy
Interaction effectsUnder-estimates governance valueWeight structural risks higher
Option valueMay over-focus on current prioritiesReserve 10-15% for exploration
Comparative advantageIgnores organizational fitApply at implementation level
Black swan risksMay miss novel risk categoriesRegular framework updates

Confidence Intervals

Estimate90% Confidence IntervalSource of Uncertainty
Misalignment share25-80%Timeline disagreement
Current allocation optimality±20 percentage pointsTractability estimates
Marginal value rankingsMedium confidenceLimited empirical data

Sources & Resources

Funding Data Sources

SourceTypeCoverageUpdate FrequencyURL
Coefficient Giving Grants DatabasePrimaryAll CG grantsReal-timeopenphilanthropy.org
EA Funds LTFF ReportsPrimaryLTFF grantsQuarterlyeffectivealtruism.org
Longview Philanthropy AnalysisAnalysisLandscape overviewAnnualEA Forum
CG Technical Safety AnalysisAnalysisCG TAIS breakdownAnnualLessWrong
Coefficient GivingAnnual reportsStrategy & prioritiesAnnualopenphilanthropy.org

Expert Surveys

SurveySampleYearKey FindingMethodology Notes
Grace et al. (AI Impacts)738 ML researchers20225-10% median x-riskNon-response bias concern
Conjecture Internal Survey22 safety researchers202380% median x-riskSelection bias (safety workers)
FLI AI Safety IndexExpert composite202524 min to midnightQualitative assessment

Academic Literature

CategoryKey PapersOrganizationRelevance
Portfolio TheoryMarkowitz (1952)University of ChicagoFoundational framework
Risk AssessmentKaplan & Garrick (1981)UCLARisk decomposition
AI Risk SurveysGrace et al. (2022)AI ImpactsExpert elicitation
MIT AI Risk RepositoryMIT2024Risk taxonomy

Policy Organizations

OrganizationFocus AreaKey Resources2024 Budget (Est.)
RAND CorporationDefense applicationsNational security risk assessments$5-10M AI-related
CSETTechnology policyAI governance frameworks$8-12M
CNASSecurity implicationsMilitary AI analysis$3-5M AI-related
Frontier Model ForumIndustry coordinationAI Safety Fund ($10M+)$10M+

This framework connects with several other analytical models:

  • Compounding Risks Analysis - How risks interact and amplify
  • Critical Uncertainties Framework - Key unknowns affecting strategy
  • Capability-Alignment Race Model - Timeline dynamics
  • AI Safety Defense in Depth Model - Multi-layered risk mitigation

References

RAND Corporation is a nonprofit research organization providing objective analysis and policy recommendations across a wide range of topics including national security, technology, governance, and emerging risks. It produces influential studies on AI policy, cybersecurity, and global governance challenges. RAND's work is frequently cited by governments and policymakers worldwide.

★★★★☆

Partnership on AI (PAI) is a nonprofit coalition of AI researchers, civil society organizations, academics, and companies working to develop best practices, conduct research, and shape policy around responsible AI development. It brings together diverse stakeholders to address challenges including safety, fairness, transparency, and the societal impacts of AI systems. PAI serves as a coordination hub for cross-sector dialogue on AI governance.

★★★☆☆

Epoch AI is a research organization focused on investigating and forecasting trends in artificial intelligence, particularly around compute, training data, and algorithmic progress. They produce empirical analyses and datasets to inform understanding of AI development trajectories and support better decision-making in AI governance and safety.

★★★★☆

The 2022 ESPAI surveyed 738 machine learning researchers (NeurIPS/ICML authors) about AI progress timelines and risks, serving as a replication and update of the 2016 survey. Key findings include an aggregate forecast of 50% chance of HLMI by 2059 (37 years from 2022), with significant disagreement among experts about timelines and risks.

★★★☆☆

The Center for a New American Security's Technology and National Security Program produces policy research and recommendations focused on U.S.-China competition in AI, biotechnology, next-generation communications, and quantum technologies. It aims to help U.S. and allied policymakers maintain technological leadership while managing risks to security and democratic values. The program bridges technology and policy communities to develop actionable governance frameworks.

★★★★☆

CNAS is a Washington D.C.-based national security think tank publishing research on defense, technology policy, economic security, and AI governance. Its Technology & National Security program produces policy-relevant work on AI, cybersecurity, and emerging technologies with implications for AI safety and governance.

★★★★☆
7AI Philanthropy's 2023 reportaiphilanthropy.org

AI Philanthropy's 2023 report surveys philanthropic funding landscapes related to artificial intelligence, likely covering grant allocations, priority areas, and strategic recommendations for donors interested in AI-related causes including safety. It provides insight into how philanthropic capital is being deployed across AI research, governance, and safety initiatives.

Good Judgment Open is a crowd-sourced forecasting platform where participants predict geopolitical, economic, and technological events, with top performers earning the 'Superforecaster' designation. Founded by Philip Tetlock, whose research demonstrated that structured probabilistic thinking can dramatically improve prediction accuracy. The platform serves as both a competitive forecasting community and a research tool for studying human judgment under uncertainty.

9Portfolio Selection (Markowitz, 1952)Wiley Online Library (peer-reviewed)·Paper

Harry Markowitz's foundational 1952 paper introduces modern portfolio theory, establishing the mathematical framework for optimal portfolio construction by balancing expected return against variance (risk). It introduces the concept of the 'efficient frontier' and demonstrates that diversification can reduce risk without sacrificing expected returns. This work formalized the trade-off between risk and return, revolutionizing financial economics and decision-making under uncertainty.

★★★★☆

RAND Corporation's AI research hub covers policy, national security, and governance implications of artificial intelligence. It aggregates reports, analyses, and commentary on AI risks, military applications, and regulatory frameworks from one of the leading U.S. defense and policy think tanks.

★★★★☆
11portfolio optimization theoryOxford Academic (peer-reviewed)

This academic paper from the Review of Financial Studies presents foundational theory on portfolio optimization, likely building on Markowitz mean-variance framework. It addresses how investors should allocate resources across assets to optimize risk-return tradeoffs under uncertainty.

★★★★★

A self-critical assessment by Open Philanthropy leadership proposing changes to the organization's cause prioritization framework, resource allocation, and strategic direction. The piece reflects on lessons learned and argues for adjustments in how the organization balances near-term vs. long-term priorities, including AI safety funding. It represents an important public accounting of how a major AI safety funder thinks about portfolio construction.

★★★★☆

80,000 Hours is a nonprofit that provides research and advice on how to use your career to have the most positive impact on the world's most pressing problems, with significant focus on AI safety and existential risk. They offer career guides, job boards, and in-depth research on high-priority cause areas and career paths. Their methodology emphasizes earning to give, direct work in high-impact fields, and building career capital.

★★★☆☆
14CSET: AI Market DynamicsCSET Georgetown

CSET (Center for Security and Emerging Technology) at Georgetown University is a policy research organization focused on the security implications of emerging technologies, particularly AI. It produces research on AI policy, workforce, geopolitics, and governance. The content could not be fully extracted, limiting detailed analysis.

★★★★☆
15Kaplan & Garrick (1981)Wiley Online Library (peer-reviewed)·Lee R. Abramson·1981·Paper

Kaplan and Garrick (1981) propose a foundational quantitative definition of risk based on the concept of a 'set of triplets,' establishing a formal framework for risk analysis. The paper extends this definition to incorporate uncertainty and completeness, utilizing Bayes' theorem for probabilistic reasoning. The authors apply their framework to discuss key concepts including relative risk, the relativity of risk across contexts, and the acceptability of risk, providing a mathematical foundation for systematic risk assessment that has become influential in the risk analysis field.

★★★★☆

Open Philanthropy's focus area page on potential risks from advanced AI outlines their strategic grantmaking approach to reducing catastrophic and existential risks from transformative AI systems. It explains their reasoning for prioritizing AI safety research, policy work, and field-building as among the most important philanthropic opportunities of our time.

★★★★☆
17AI ImpactsAI Impacts

AI Impacts is a research organization that investigates empirical questions relevant to AI forecasting and safety, including AI timelines, discontinuous progress risks, and existential risk arguments. It maintains a wiki and blog featuring expert surveys, historical analyses, and structured arguments about transformative AI development. Notable outputs include periodic expert surveys on AI progress timelines.

★★★☆☆
18FLI AI Safety Index Summer 2025Future of Life Institute

The Future of Life Institute's AI Safety Index Summer 2025 systematically evaluates leading AI companies on safety practices, finding widespread deficiencies across risk management, transparency, and existential safety planning. Anthropic receives the highest grade of C+, indicating that even the best-performing company falls significantly short of adequate safety standards. The report serves as a comparative benchmark for industry accountability.

★★★☆☆

The AI Safety Fund (AISF) is a $10 million+ collaborative initiative launched in October 2023 by Anthropic, Google, Microsoft, and OpenAI (via the Frontier Model Forum) along with philanthropic partners to fund independent AI safety and security research. It has distributed two rounds of grants focused on responsible frontier AI development, public safety risk reduction, and standardized third-party capability evaluations. The fund is now directly managed by the Frontier Model Forum following the closure of its original administrator, the Meridian Institute.

★★★☆☆

Related Wiki Pages

Top Related Pages

Approaches

AI Safety Intervention Portfolio

Analysis

AI Safety Defense in Depth ModelAI Risk Interaction MatrixCapability-Alignment Race ModelAI Safety Research Value ModelCapabilities-to-Safety Pipeline ModelAI Safety Research Allocation Model

Key Debates

AI Risk Critical Uncertainties Model

Organizations

AnthropicCoefficient GivingMachine Intelligence Research InstituteCenter for AI SafetyGovAIGeorgetown CSET

Concepts

AGI Timeline

Other

Jaan Tallinn