Global Catastrophic Risk Institute

Theory of Change

GCRI was founded in 2011 to be the integrator across global catastrophic risks -- the org that studies nuclear, AI, bio, climate, and asteroid risks together instead of in silos. Their stated theory: siloed risk work "only goes so far, and indeed it can even be counterproductive by increasing one risk while reducing another." GCRI exists to find cross-risk synergies and tradeoffs.

On AI specifically, Baum's 2025 position is heterodox: he views current LLMs as "bags of heuristics" unlikely to achieve AGI, and advocates steering the AI industry toward "products not gods" rather than pursuing extreme AGI. He argues for monitoring warning signs, contingency planning, and international cooperation. GCRI's AI work focuses on governance (corporate governance mechanisms, collective action between AI groups, certification) and risk analysis (LLM takeover assessment, ASI-PATH fault tree) rather than technical alignment.

GCRI's broader strategic framework emphasizes "win-win solutions" -- designing interventions that reduce GCR while also serving stakeholders' existing priorities, so that non-GCR-focused decision-makers will actually implement them.

What They Do

Research output: ~60+ publications since 2011 across AI, nuclear, climate, ethics, space, and cross-risk topics. Published in Risk Analysis, AI & Society, Technology in Society, IEEE, Natural Hazards, Futures, Science, BBC Future, Bulletin of the Atomic Scientists, Project Syndicate. Peak: 9 publications in 2022. In 2024: 5 publications (LLM takeover in Risk Analysis, COVID origin survey, climate/GCR paper, diversity ethics, social choice manipulation).

Notable AI publications: Corporate governance of AI in the public interest (2021, with Jonas Schuett), LLM takeover risk assessment (2025, Risk Analysis), AGI project surveys (2017, 2020), collective action on AI (2021), AI certification (2021, IEEE).

Programs: Annual Advising and Collaboration Program (23-81 advisees/year), Fellowship Program (3-12 fellows/year). Policy engagement through IEEE P2863 AI governance standard, NIST AI Risk Management Framework, OSTP submissions.

Recent pivot (2025-2026) toward public outreach: new website, YouTube channels, GCRI Symposium on World Peace, commentary on Iran war and Ukraine nuclear risk.

Key People

Seth Baum -- Co-Founder and Executive Director. Sole active staff since late 2024. Ph.D. Geography (Penn State), M.S. Electrical Engineering (Northeastern). CSER Research Affiliate, Science and Engineering Ethics associate editor. The most prolific author in the GCR/ER Foundations cluster per 2025 bibliometric review (4,113 Google Scholar citations). Unusually broad intellectual range: AI governance, nuclear risk modeling, climate, asteroid deflection, space ethics.

Tony Barrett -- Co-Founder, Director of Research 2011-2024, now Senior Advisor. Departed to DHS CISA National Risk Management Center (Jan 2025). Former RAND Stanton Nuclear Security Fellow. His government placement is arguably GCRI's highest-impact policy outcome.

Staff trajectory: peak of 5 (2021-2022), then steady attrition with no replacements. De Neufville departed 2021, Fitzgerald and Owe departed late 2022, Barrett departed late 2024. Baum in the 2024 annual report: "I take full responsibility for this."

Money and Incentives

Total lifetime funding: ~$2M over 14 years (~$143K/year average).

Revenue breakdown:

• Gordon Irlam (via charitable foundation): ~$1.35M (2016-2022), specifically for AI work. Provided ~67% of all known funding. No evidence of continued funding after 2022.
• SFF/Jaan Tallinn: $487K (2019-2021). ~24% of all known funding. No grants after 2021.
• FLI: ~$100K (2015, ASI-PATH project).
• Other: BERI $25K, EA Grants $24K, Donor Lottery $20K.

Notable absences: Zero Coefficient Giving/Open Phil grants (across 480+ AI safety grants). No government grants. No foundation grants outside the EA ecosystem.

Budget trajectory: ~$140K (2018) -> ~$250K (2019) -> ~$400K (2021 peak) -> ~$350K (2022) -> unknown (2023-present). Reserves in 2022 were sufficient "through early 2024." No budget disclosed since.

Business model: Grants and individual donations, processed through fiscal sponsor SEE. No product revenue, no consulting income, no endowment. GCRI considered disbanding in 2024.

Incentive analysis: GCRI has no AI lab funding, no corporate sponsors, no compute dependencies, no board conflicts. This is the cleanest incentive structure in AI safety -- entirely because they are the poorest org in the space. The flip side: zero financial resilience means the org's existence depends on finding new funders willing to support cross-risk work or Baum's continued willingness to operate at presumably minimal compensation.

What Others Say

Adam Gleave (2018, EA donor lottery review): "Overall I am moderately excited about supporting the work of GCRI and in particular Seth Baum. I am pessimistic about room for growth, with recruitment being a major challenge."

Brookings scholars (2022): Engaged substantively with Baum's nuclear probability methodology. Agreed frequentist approaches fail, but argued even Bayesian estimates are "useful for thinking about adjusting one's own subjective estimate of probability, but not for informing the actual probability of Putin deciding to use a nuclear weapon." Recommended possibilistic approaches instead.

Bibliometric review (Earth System Dynamics, 2025): Identified Baum as the most prolific author in the GCR/ER Foundations cluster. GCRI listed alongside FHI, FLI, CSER as a "key founding institution." But the review also noted the field is dominated by "a small subset of authors."

EA community: Essentially invisible. Zero posts on LessWrong, EA Forum, or Alignment Forum. No Coefficient Giving funding. GCRI operates in academic and policy circles, not the EA ecosystem.

What's Absent

• No financial disclosure for 2023 or 2024. Current budget and Baum's compensation are entirely opaque.
• No impact metrics of any kind -- no policy outcomes, citation analysis, or advisee tracking after 14 years.
• No independent governance structure. No board. Baum has sole operational authority under fiscal sponsor umbrella.
• No technical AI safety research. All work is governance, ethics, and risk analysis.
• No collaboration with any frontier AI lab.
• No explanation of why three staff members departed in 2021-2022, or why none were replaced.
• No succession plan. Bus factor of 1.

Recommended Reading

1. FLI Podcast: "What Are the Odds of Nuclear War?" (2022) -- Baum and de Neufville discuss methodology, uncertainty, and nuclear risk with genuine candor. Best window into how GCRI actually thinks about problems. futureoflife.org/podcast/podcast-what-are-the-odds-of-nuclear-war-a-conversation-with-seth-baum-and-robert-deneufville/

2. "Parsing AI Risk in Early 2025" -- Baum's most current and candid AI views: LLMs as bags of heuristics, products not gods, skepticism of imminent AGI. A heterodox position within AI safety. gcri.org/publications/commentary/parsing-ai-early-2025

3. Brookings: "How Not to Estimate the Likelihood of Nuclear War" (2022) -- The strongest external engagement with GCRI's methodology. A respectful but substantive critique. brookings.edu/articles/how-not-to-estimate-the-likelihood-of-nuclear-war/

4. 2024 Annual Report -- Baum's admission of sole-operator status, consideration of disbanding, and vision for GCRI's future. The most revealing organizational document. gcri.org/about/annual-report/2024