The RAND Corporation is a US nonpartisan policy research nonprofit that originated as a US Air Force operations-research project and now produces analytical research for policymakers across national security, health, education, labor, and technology. In AI policy it is a producer of quantitative governance research, particularly on the compute, power, security, and national-security dimensions of frontier AI, and is the current institutional home of compute-governance researcher Lennart Heim. Its AI work is concentrated in the Technology and Security Policy Center (TASP) within RAND Global and Emerging Risks (RAND — How AGI Could Affect the Rise and Fall of Nations; Source: rand.org).
| Field | Value |
|---|---|
| Type | 501(c)(3) policy research nonprofit (originated as a US Air Force project, 1948) |
| Headquarters | Santa Monica, California |
| Founded | 1948 (spun out of Project RAND at Douglas Aircraft) |
| President / CEO | Jason Matheny (since July 2022; previously founding director of CSET and director of IARPA) (Source: rand.org) |
| AI research unit | Technology and Security Policy Center, RAND Global and Emerging Risks |
| Known for | Defense analysis, nuclear strategy, systems analysis, public policy, and AI/compute governance research |
Overview
RAND produces nonpartisan, analytical research for policymakers across national security, health, education, labor, and technology. It originated as an Air Force operations-research project during the early Cold War, spun out of Project RAND at Douglas Aircraft in 1948, and retains ties to the US defense and intelligence communities; its research now spans dozens of domestic and international policy domains.
RAND's association with artificial intelligence predates the modern field's name. Allen Newell, Herbert Simon and Cliff Shaw of RAND developed the Logic Theorist in 1955–56, generally described as the first running artificial-intelligence program, and presented it at the 1956 Dartmouth workshop (Herbert A. Simon).
AI policy work
RAND is an institutional home for quantitative AI governance research, concentrated on the compute, power, security, and national-security dimensions of frontier AI. Its prominence in this area is associated with three factors. Jason Matheny, who founded CSET at Georgetown, directed the Intelligence Advanced Research Projects Activity (IARPA), and worked on technology and national-security policy at the National Security Council and the Office of Science and Technology Policy, became RAND's president and CEO in July 2022 (Source: rand.org). Lennart Heim moved from GovAI (Oxford) to RAND, where he leads compute research in the Technology and Security Policy Center and holds a professorship at the Pardee RAND Graduate School, anchoring the institution's compute and power forecasting work (Source: rand.org). RAND's traditional US government client base (DoD, DOE, the intelligence community) gives its AI work direct policy uptake.
RAND also publishes commissioned work on AI national security, autonomy, and military AI procurement that is referenced across CSIS and CSET analyses, and produced dual-use biology studies in 2023–2024 on the biosecurity implications of large language models (CBRN Uplift).
Research themes
- Compute governance — training-compute measurement, power forecasting, verification infrastructure (Compute Governance)
- Model-weight security — threat modelling and security levels for frontier-model weights (below)
- Export controls — analysis of BIS controls' effectiveness and design, including hardware-enabled governance mechanisms and the AI diffusion framework (Export Controls (AI); Lennart Heim)
- AI and national security — DoD adoption, autonomy, military competition with China, AGI scenarios
- AI policy design — frameworks for agency AI governance, risk management, procurement, and historical analogues for governance (Historical Analogues That Can Inform AI Governance — Michael J. D. Vermeer (RAND, 2024))
Key publications
AI's Power Requirements Under Exponential Growth (Pilz, Mahmood, Heim, RAND RRA3572-1, 2025) is a reference forecast used in US federal energy-AI planning (RAND — AI's Power Requirements Under Exponential Growth (2025)). It estimates US AI data-center power at 5–8 GW in 2024, rising to 40–80 GW by 2030, potentially 8–15% of US electricity consumption, with a central single-run estimate of roughly 2–10 GW, and argues that power, not chips, is becoming the binding constraint on AI scaling. RAND positions the report as a policy-facing companion to Epoch AI's compute forecasts (Epoch AI — How Much Power Will Frontier AI Training Demand in 2030?); it has been cited by CSIS, the Department of Defense, and BIS.
Securing AI Model Weights: Preventing Theft and Misuse of Frontier Models (Nevo, Lahav, Karpur, Bar-On, Bradley, Alstott, RAND RR-A2849-1, May 30, 2024) examines what it would take to protect the weights of frontier models from theft. It identifies 38 "meaningfully distinct attack vectors" and defines five security levels corresponding to threat actors "from opportunistic (often financially driven) criminals to highly resourced nation-state operations" (Source: rand.org). Its SL2–SL5 scale has become a shared reference point for lab security: the AI 2027 scenario describes a fictional frontier lab as operating at "RAND's SL2, working toward SL3", with nation-state-level defense (SL4–5) still distant.
How AGI Could Affect the Rise and Fall of Nations (Pavel, Ke, Smith, Brown-Heidenreich, Sabbag, Acharya, Mahmood, RAND RR-A3034-2, July 2025) presents eight illustrative scenarios at the intersection of AGI-development centralization and geopolitical power shifts, applying RAND's assumption-based planning method (RAND — How AGI Could Affect the Rise and Fall of Nations).
Historical Analogues That Can Inform AI Governance (Michael J. D. Vermeer, RAND, August 2024) draws on nuclear, internet, biotechnology, chemical-weapons and financial-regulation precedents to inform AI governance design (Historical Analogues That Can Inform AI Governance — Michael J. D. Vermeer (RAND, 2024)).
Heim's RAND work also includes Understanding the AI Diffusion Framework: Can Export Controls Create a U.S.-Led Global Artificial Intelligence Ecosystem? (Perspective PEA3776-1, January 2025) and the working paper Hardware-Enabled Governance Mechanisms (WRA3056-1, January 2024, with Kulp, Puri, Gonzales, Vermeer, Smith and Winkelman) (Lennart Heim).
RAND wargaming on autonomous systems is cited in reporting on the AI arms race: exercises found that "the speed of autonomous systems led to inadvertent escalation" (Mutually Automated Destruction: The Escalating Global A.I. Arms Race (NYT)).
Funding and governance
RAND is primarily funded through federal contracts and grants (DoD, HHS, DHS, and other agencies), supplemented by foundation grants and philanthropic gifts. Its federally-funded research-and-development-center (FFRDC) work for the Air Force, Army, and intelligence community is long-standing. AI-specific funding includes contracts from NIST, the National Science Foundation, and defense clients, as well as foundation grants from Open Philanthropy and Hewlett for governance research.
Key people
- Jason Matheny — President and CEO since July 2022; founding director of CSET; former IARPA director; NSC and OSTP technology and national-security roles; appointed by Congress in 2018 to the National Security Commission on Artificial Intelligence (Source: rand.org).
- Lennart Heim — leads compute research in the Technology and Security Policy Center; co-author of the AI power-requirements report.
- Michael J. D. Vermeer — author of the historical-analogues report and co-author of the hardware-enabled governance working paper.
- Sella Nevo — lead author of the model-weights security report.
- Additional RAND researchers contribute to AI work; the institution's output is large and distributed. Former RAND staff appear across the AI-policy ecosystem; the Center for AI Safety, for example, records a computational social scientist formerly at RAND among its staff.
Relationships
- related: Lennart Heim — current institutional affiliation; anchors RAND compute work
- related: Center for Security and Emerging Technology (CSET) — Matheny's prior institution; parallel natsec-AI research shop
- related: Center for Strategic and International Studies (CSIS) — parallel think tank; RAND data feeds CSIS policy writing
- related: Brookings Institution AI Initiative — peer institution with a domestic-policy framing where RAND's is national-security
- related: Compute Governance, AI Environmental Impact, Semiconductor Supply Chain
- related: Export Controls (AI), AI Race Dynamics
- related: Herbert A. Simon — Logic Theorist developed at RAND, 1955–56
- supports: RAND — AI's Power Requirements Under Exponential Growth (2025)
- supports: RAND — How AGI Could Affect the Rise and Fall of Nations
- supports: Historical Analogues That Can Inform AI Governance — Michael J. D. Vermeer (RAND, 2024)
- related: AI 2027 — the scenario adopts RAND's SL2–SL5 weight-security levels