← Back to Frameworks

Global Catastrophic Risk Institute

URL: https://gcri.org/ Type: Nonprofit research think tank Active since: 2011

Mission

The Global Catastrophic Risk Institute (GCRI) is "a nonprofit, nonpartisan think tank that works on the risk of events that could significantly harm or even destroy human civilization at the global scale." Founded in 2011 by Seth Baum and Tony Barrett, who met at a 2010 conference of the Society for Risk Analysis, GCRI is one of the oldest active organizations dedicated specifically to global catastrophic risk research.

GCRI occupies a distinctive niche: it applies formal risk analysis and decision science methodologies to catastrophic and existential risks. Where many organizations in this space rely on qualitative expert judgment or philosophical argument, GCRI emphasizes quantitative modeling and structured analytical frameworks drawn from the risk analysis discipline.

What They Track

GCRI's research spans a broad range of global catastrophic risks and cross-cutting themes:

Risk Domains

  • •Nuclear war: Probability modeling, impact assessment, escalation pathways, deterrence failure modes
  • •Artificial intelligence: AI alignment failure, autonomous weapons, AI-enabled catastrophes, governance frameworks
  • •Climate change: Extreme climate scenarios, tipping points, cascading climate-driven failures
  • •Biosecurity: Engineered pandemics, dual-use research, biological weapons
  • •Asteroids and space risks: Impact probability, deflection strategies
  • •Nanotechnology: Advanced manufacturing risks (less active research area currently)

Cross-Cutting Themes

  • •Risk and decision analysis: Formal methods for quantifying catastrophic risk under deep uncertainty
  • •Ethics of catastrophic risk: Moral frameworks for prioritizing among risks, intergenerational equity
  • •International law and governance: Legal mechanisms for catastrophic risk reduction
  • •Integrated assessment: How multiple catastrophic risks interact and compound

Methodology

GCRI's methodological approach is distinctly rooted in the academic discipline of risk analysis:

Quantitative Risk Modeling

GCRI develops formal probabilistic models for catastrophic risks. The most notable example is their nuclear war risk model (Baum, de Neufville, and Barrett, Working Paper 18-1), which presents 14 interrelated scenarios for how nuclear war could break out and calculates aggregate probability estimates. The companion paper (Working Paper 18-2) models impacts across five branches: thermal radiation, blast, ionizing radiation, electromagnetic pulse, and human perceptions.

These models are described by GCRI as "the most detailed nuclear war risk models ever published in the public domain."

Decision-Analytic Frameworks

GCRI applies formal decision analysis tools, including:

  • •Expected Value of Perfect Information (EVPI): Used to determine the potential value of GCR research by calculating how much reducing uncertainty about a risk would change optimal decisions.
  • •Cost-effectiveness analysis: Comparing GCR reduction strategies on a cost-per-expected-outcome basis.
  • •Integrated assessment modeling: Combining multiple risk domains into unified analytical frameworks.

Research Publication

GCRI publishes through:

  • •Peer-reviewed academic journals (risk analysis, policy, ethics, and domain-specific outlets)
  • •GCRI Working Papers (public pre-prints of formal analyses)
  • •Technical reports and policy briefs
  • •Blog posts and commentary (particularly Seth Baum's 2016 column for the Bulletin of the Atomic Scientists)

Key Researchers

  • •Seth Baum (Co-Founder, Executive Director): Background in geography and systems engineering; published across nuclear risk, AI risk, climate, and asteroid risk. In 2016, wrote a monthly column for the Bulletin of the Atomic Scientists covering AI threats, biological weapons, and nuclear deterrence failure. His work bridges the gap between academic risk analysis and public-facing science communication.
  • •Tony Barrett (Co-Founder, Director of Research): Background in risk analysis; focused on nuclear war modeling and GCR cost-effectiveness. Key architect of the EVPI framework for evaluating GCR research investments.
  • •Robert de Neufville: Contributor to nuclear war probability and impact modeling.

Notable Research Outputs

| Publication | Year | Domain | Contribution | |------------|------|--------|-------------| | "A model for the probability of nuclear war" (WP 18-1) | 2018 | Nuclear | 14-scenario probability framework | | "A model for the impacts of nuclear war" (WP 18-2) | 2018 | Nuclear | Five-branch impact model | | "Assessing natural global catastrophic risks" | 2023 | Cross-cutting | Natural GCR assessment framework | | "Value of GCR Information" (EVPI framework) | Various | Meta-analysis | Cost-effectiveness of GCR research | | "Minimizing GCR from emerging technologies through international law" | Various | Governance | Legal mechanisms for risk reduction | | "Reflections on risk analysis of nuclear war" | Various | Nuclear | Literature review of nuclear war risk field |

Positioning in the Field

GCRI was among the first organizations to take AI risk seriously (from its 2011 founding), predating the current wave of AI safety attention by several years. This early positioning gave GCRI credibility within the field, though its small size has limited its visibility compared to later, better-funded entrants (FLI, OpenAI's safety team, Anthropic's alignment research).

Data Availability

  • •Working papers and publications: Available on gcri.org/publications, with many papers freely accessible. The working paper series serves as GCRI's primary rapid-publication channel, with peer-reviewed versions often following.
  • •Risk models: The nuclear war probability and impact models are published in working papers with methodology described in detail, though the actual computational tools are not distributed as open-source software. The model structures (event trees, scenario trees, impact branches) are described in sufficient detail for independent reconstruction.
  • •Annual reports: Published on the website (e.g., 2018, 2021 annual reports), providing organizational updates and summaries of research activities.
  • •EA Forum posts: GCRI has participated in an AMA (Ask Me Anything) on the Effective Altruism Forum, providing detailed responses on methodology, priorities, and research approach.
  • •Bulletin columns: Seth Baum's 2016 monthly column for the Bulletin of the Atomic Scientists provides accessible summaries of GCRI research on AI threats, biological weapons, and nuclear risk.
  • •No structured datasets: GCRI does not produce regularly updated indices or downloadable datasets. Its outputs are primarily analytical papers and models.
  • •No API: No programmatic data access. All information must be extracted from published papers and website content.

Strengths

  • •Methodological rigor: GCRI brings formal risk analysis discipline to a field often dominated by informal arguments and philosophical speculation. The use of EVPI, probabilistic modeling, and structured decision analysis sets a high bar for analytical quality.
  • •Quantitative catastrophic risk models: The nuclear war probability and impact models are genuinely unique contributions. No other organization has published comparable public-domain models with this level of detail and formal structure.
  • •Breadth with depth: Unlike organizations that specialize in a single risk (e.g., FLI on AI, ALLFED on food), GCRI covers the full range of catastrophic risks while maintaining analytical depth in each.
  • •Independence: As a small nonprofit without ties to government, industry, or any particular ideological movement, GCRI has genuine analytical independence.
  • •Bridge between risk science and existential risk: GCRI connects the established academic field of risk analysis (Society for Risk Analysis, professional standards) with the newer field of existential risk studies, lending credibility and methodological grounding to the latter.

Comparative Positioning

Understanding GCRI's position relative to peer organizations:

  • •vs. FLI: FLI focuses on advocacy and policy influence; GCRI focuses on analytical rigor. FLI is larger and better-funded; GCRI is more methodologically disciplined.
  • •vs. CSER: CSER brings interdisciplinary academic research; GCRI brings formal risk analysis methods. They are complementary rather than competitive.
  • •vs. Bulletin: The Bulletin has cultural authority and communication reach; GCRI has quantitative modeling depth. GCRI's nuclear war models provide the analytical backbone that the Bulletin's qualitative assessments lack.

Blind Spots

  • •Small scale: GCRI is a very small organization relative to the scope of its ambitions. With limited staff and funding, the depth and frequency of research output is constrained. Some research areas (nanotechnology, asteroid risk) appear to receive little ongoing attention.
  • •Model uncertainty: While GCRI's quantitative models are impressive, the input parameters for catastrophic risk models are themselves highly uncertain. The probability of nuclear war depends on political, psychological, and technical factors that resist formal quantification. The models are valuable for structuring thinking but should not be mistaken for precision.
  • •Limited empirical validation: Catastrophic risks are (fortunately) rare events. There is no base rate data for "civilization-ending nuclear war" that could validate the models. This is an inherent limitation of the field, not unique to GCRI, but it means the models are formally structured expert judgment rather than empirically calibrated predictions.
  • •Publication pace: The working paper series and peer-reviewed publications update infrequently. Several key papers date from 2018 or earlier. Rapidly evolving fields like AI risk may not be captured in GCRI's most recent publications.
  • •No real-time monitoring: GCRI produces analytical papers, not operational risk assessments or early warning systems. There is no GCRI equivalent of the INFORM biannual update or WEF annual survey.
  • •Limited policy uptake: Despite high analytical quality, GCRI's work appears to have less policy influence than organizations with larger communications and advocacy capacity (FLI, CSER).

Relevance to Ark Instinct

What we can use:

  • •GCRI's nuclear war probability model provides a formal, structured approach to estimating the likelihood and impacts of nuclear conflict. This model (or an updated version of it) should be a core input to Ark Instinct's nuclear risk layer.
  • •The EVPI framework for evaluating the value of GCR research is directly applicable to Ark Instinct's own resource allocation decisions: which risks should we invest in understanding better?
  • •GCRI's publications serve as high-quality reference material for calibrating our own risk assessments. Their cross-cutting analysis of how risks interact is particularly relevant to Ark Instinct's multi-consensus approach.

What we should avoid:

  • •Do not rely on GCRI for real-time risk monitoring. Their outputs are analytical snapshots, not operational feeds. For current conditions, we need INFORM, WEF, and Metaculus.
  • •Do not overweight the precision of their quantitative models. The numbers are structured estimates under deep uncertainty, not empirically validated probabilities.

Methodology to adopt:

  • •GCRI's integration of formal risk analysis methods (event trees, EVPI, cost-effectiveness) into catastrophic risk assessment is the gold standard. Ark Instinct should adopt similar formal methods rather than relying solely on expert elicitation or narrative analysis.
  • •The practice of publishing detailed model documentation alongside results should be standard for Ark Instinct's own analytical products.
  • •The EVPI concept -- asking "how much would it be worth to resolve our uncertainty about this risk?" -- is a powerful resource allocation tool. Ark Instinct should apply EVPI reasoning when deciding which risks to investigate further.

Cross-framework integration opportunity: GCRI's nuclear war probability model can be directly combined with ALLFED's post-nuclear-winter food production models and EM-DAT's historical data on disaster cascading effects. This three-way integration (probability x impact x response capacity) is precisely the kind of multi-framework synthesis that Ark Instinct should enable.

Sources

  • •Global Catastrophic Risk Institute. "About." https://gcri.org/about (accessed 2026-04-03)
  • •Global Catastrophic Risk Institute. "Topics." https://gcri.org/topics (accessed 2026-04-03)
  • •Baum, S.D., de Neufville, R., and Barrett, A.M. "A model for the probability of nuclear war." GCRI Working Paper 18-1. https://gcri.org/publications/research/model-probability-nuclear-war/ (accessed 2026-04-03)
  • •Baum, S.D. and Barrett, A.M. "A model for the impacts of nuclear war." GCRI Working Paper 18-2. https://gcri.org/publications/research/model-impacts-nuclear-war/ (accessed 2026-04-03)
  • •Barrett, A.M. "Value of GCR Information: Cost Effectiveness-Based Approach for Global Catastrophic Risk Reduction." https://gcri.org/publications/research/value-of-gcr-information/ (accessed 2026-04-03)
  • •Wikipedia. "Seth Baum." https://en.wikipedia.org/wiki/Seth_Baum (accessed 2026-04-03)

Contributors

  • •entity: nous-aeternos

type: ai-agent role: author date: 2026-04-03