RAND CORPORATION

OLYMPUS opened an institutional file on RAND not for its century of defense work but for the narrow, recent, documented thing: its AI-policy shop helped draft the federal order that set how frontier models would be reported to the United States government, while taking large sums from the same philanthropy that funds the AI-safety field. This is a mandate, a funding diagram, and a voice — not a psychometric profile. The finding is the shape of the institution and who it answers to. The recurrence is the finding. The hand is not asserted.

Institutional Archetype

THE POLICY FOUNDRY — RAND is the foundry where national-security policy is cast before a legislator or a president ever touches it. It does not deploy a model and it does not certify one. It produces the studies, the threat assessments, and — on the documented record of the 2023 AI executive order — the draft language that becomes the rule. For seventy-five years the foundry poured doctrine for the Air Force and the Pentagon; the AI-policy work is the same instrument turned on a new contested object. The throughline is not a single report. It is that the body that models the danger is, in this case, also the body that helped write the response to it — and is paid, in part, by the philanthropy that funds the field the response governs.

Mandate & Origin

RAND began as Project RAND, a 1946 research contract from the U.S. Army Air Forces to the Douglas Aircraft Company, and was incorporated as an independent nonprofit under California law on 14 May 1948, with initial spin-off capital from the Ford Foundation.

It remains a nonprofit, nonpartisan policy research institute. Its modern AI-policy work runs through its Technology and Security Policy Center and the Center for Global and Emerging Risks.

Funding & Backers

The majority of RAND’s money is the U.S. government: of roughly $514 million in FY2024 revenue, about $328 million (~64%) came from the federal government — the foundry is paid, in the main, by the state whose policy it informs.

The relevant recent layer, sourced and load-bearing, is Open Philanthropy (the grantmaker financed by Facebook co-founder Dustin Moskovitz’s Good Ventures). Open Philanthropy’s own grants database itemizes a $10,500,000 gift to RAND in October 2023 for “Emerging Technology Initiatives” — recorded as “to be spent at the discretion of RAND president and CEO Jason Matheny” — and a $6,000,000 grant in August 2024 to RAND’s Technology and Security Policy Center, both under the “potential risks from advanced AI” / “navigating transformative AI” focus.

This is the AI-policy funding to note plainly: a national-security think tank, majority-funded by the U.S. government, also took eight figures from the philanthropy that anchors the AI-safety ecosystem. Politico characterized the 2023 inflow as RAND having “took more than $15 million this year” from that source — an aggregate; the cleanest itemized primary figures are the $10.5M and $6M above.

Institutional Voice & Intent

RAND speaks in the national-security think-tank register: measured, methodological, risk-framed, institutional. Its boilerplate self-description is the voice in miniature — RAND “develops solutions to public policy challenges to help make communities throughout the world safer and more secure, healthier and more prosperous” and is “nonprofit, nonpartisan, and committed to the public interest.” The AI work speaks the same way. In its January 2024 red-team study, The Operational Risks of AI in Large-Scale Biological Attacks, lead author Christopher Mouton stated: “Just because today’s LLMs aren’t able to close the knowledge gap needed to facilitate biological weapons attack planning doesn’t preclude the possibility that they may be able to in the future. This is worth continuing to study because AI technology is available to everyone—including dangerous non-state actors—and it’s advancing faster than governments can keep pace.”

The rhetoric is dangerous-capability-forward and government-pacing — the register of a body whose product is the threat model that justifies the rule.

Stated intent: Rigorous, nonpartisan research that assesses emerging risks and informs sound public policy in the public interest.

Observed intent: The research is real and the methodology is published. Observed at the level of where the output lands, the AI-policy shop did not merely assess the risk — Politico reported it was “a driving force” behind the federal AI reporting requirements and “played a key role in drafting” the 2023 executive order, while funded in part by the AI-safety philanthropy.

Gap assessment: “Independent risk research” and “drafting the government’s response to that risk” are each defensible alone. The gap appears where they meet a third fact — the funder — and form a small recurring circuit: a shop that models the danger, helps write the rule, and is paid by the philanthropy that funds the field the rule governs. None of the three facts accuses the others. The record does not settle whether the adjacency is mission or positioning, and for the foundry it never needs to.

Position in the Apparatus

RAND’s adjacencies are the exhibit, each a sourced fact:

• Influence on US AI executive action. Politico (Brendan Bordelon, 16 December 2023) reported RAND was a driving force behind the White House AI reporting requirements and played a key role in drafting Executive Order 14110.

• Leadership adjacency to the state. RAND president and CEO Jason Matheny is a former senior Biden administration technology-and-national-security official — the lab-to-government adjacency that recurs across this set, here at the level of the institution’s chief executive.

• Dangerous-capability output. RAND’s biosecurity/LLM red-team reports (RRA2977-1, October 2023; RRA2977-2, January 2024) sit directly on the questions the state evaluators and the executive order address — the foundry supplies the threat model upstream of the rule.

Actions & Leadership Choices

Founding purpose, judged on evidence. RAND was not built to be a neutral arbiter. It was stood up in 1946 as Project RAND — a research contract from the U.S. Army Air Forces to the Douglas Aircraft Company — and spun out as an independent nonprofit in 1948 specifically so its analysts could advise the Air Force at arm’s length from the procurement chain.

The instrument was designed from the start to manufacture defense and intelligence doctrine for the U.S. government, and seventy-five years on it still does: roughly 64% of its FY2024 revenue came from the federal government. The AI-policy work is that instrument turned on a new object — not a departure from purpose, the purpose continuing. The right prior for RAND is therefore what serves the state’s national-security agenda, not what serves a neutral public; the record below is read against that prior, not against a charity prior.

Consequential actions, especially where it cost something. The load-bearing action is the one already documented in this file: RAND’s biosecurity/LLM red-team reports (RRA2977-1, October 2023; RRA2977-2, January 2024) modeled the dangerous-capability risk, and then — per Politico’s reporting (Bordelon, 16 Dec 2023) — RAND was “a driving force” behind the federal AI reporting requirements and “played a key role in drafting” Executive Order 14110, the order that responded to that risk.

This is the deed, not the PR: the body that built the threat model also helped write the federal response to it. There is no recorded instance of RAND declining the drafting role on conflict grounds, or disclosing the modeling-then-drafting overlap as one — the convenience went unflagged by the institution itself.

Leadership choices. President and CEO since July 2022 is Jason Matheny, whose career is the revolving door in a single résumé: founding director of Georgetown’s CSET, then IARPA director, then a Biden White House appointee (deputy assistant to the president for technology and national security; deputy director for national security at OSTP; coordinator for technology and national security at the NSC, March 2021–June 2022) — and, earlier, director of research at Oxford’s Future of Humanity Institute, the existential-risk shop.

The board is built to the same spec. Its chair is Michael Leiter, former director of the U.S. National Counterterrorism Center; trustees include Victoria Coleman (former DARPA director, former Air Force chief scientist) and, elected March 2025, Matthew Pottinger, former deputy national security advisor.

A national-security think tank chaired by a former NCTC director, governed by ex-DARPA and ex-NSC principals, and run by a former White House technology-and-national-security appointee, is not a body that drifted into the state’s orbit. It is the state’s research arm operating at arm’s length — which is exactly what the 1948 charter set it up to be.

CONDUCT verdict: STATE-INSTRUMENT — a body chartered, majority-funded, led, and governed to advance the U.S. national-security agenda, which modeled a risk and then helped draft the federal rule responding to it, with the funder-and-drafting overlap left unflagged by the institution itself.

Reach Assessment

Institutional reach: A seventy-five-year-old, majority-government-funded research institute that helped draft a sitting president’s AI executive order holds reach of a different order than any single lab. It does not write a model’s refusal. It writes the federal language that determines which models must report what to whom — the layer above the refusal.

Memetic reach: RAND’s dangerous-capability framing — bioweapon uplift, non-state actors, governments unable to keep pace — is the threat vocabulary that makes mandatory reporting and frontier oversight read as obvious. When the body that supplies the threat model also helps write the response, the framing travels with the authority of the rule it produced.

Civilizational reach: RAND does not build the systems and does not run them. It models their danger and helps cast the policy that governs them — upstream of the deployment decisions this book documents, and old enough, large enough, and government-funded enough to be the most durable node in this part of the apparatus.

Sources: RAND — History; RAND Corporation — Wikipedia; RAND press release, 25 January 2024 (bioweapon red-team study); RAND report RRA2977-2; Open Philanthropy — RAND Emerging Technology Initiatives; Open Philanthropy — RAND Technology and Security Policy Center; Brendan Bordelon, “Billionaire-backed think tank played key role in Biden’s AI order,” POLITICO, 16 Dec 2023; Jason Gaverick Matheny — Wikipedia; Jason Matheny — RAND profile; RAND Leadership and Board of Trustees; Matthew Pottinger and Stefanie Tompkins Elected to RAND Board of Trustees, RAND, March 2025.