Our Work

The programs that defend, adopt, and advance.

Our work takes the national security problem AI poses, and brings together people from government, industry, and academia to work it.

01Defend

Model Security

Model Security protects the frontier AI the United States depends on from theft, disruption, and adversarial attack. The infrastructure it runs on, the model weights, and the people who build them are all targets that must be protected, while the technology sits at the front lines of strategic competition.

We work with the labs and with government. With the labs, we develop the technical, legal, and operational means to harden models against exfiltration and insider threats, and to close the gaps in what defenders can see. With government, we identify the authorities already available to act, and where they fall short.

In the near term, the work is practical: put existing tools and authorities to use, and fill the information gaps that leave both the labs and the government working from an incomplete picture. Over the longer term, it is structural: the technical and governance foundations that keep American AI secure and ahead as the technology matures.

02Adopt

AI Adoption & Evaluation

Commercial benchmarks were not built for national security. The government has needs the market does not measure: reliability in conditions where failure is not an option, security requirements of its own, and missions where a wrong answer is measured in lives or strategic position. AI Adoption & Evaluation builds the criteria that separate a system ready for that from one that is not.

The work has two parts.

Capability awareness

An independent way to see what is emerging from the labs and to understand what it means for the mission before it is fielded or released.

Test & evaluation

The frameworks that determine whether an AI-enabled system is fit for a specific government use, built to fit the acquisition and testing processes the government already runs.

The aim is confidence on both sides. Developers and government users come to understand each other’s needs, acquisition and operational risk goes down, and the systems that reach the field are the ones that let critical missions proceed with confidence, from the tactical level to the strategic.

03Advance

U.S. Capability for Global AI Governance

As AI capability accelerates, it shifts the balance of power between states, faster than doctrine and diplomacy can adjust. This program studies how the United States can turn its lead in AI into lasting strategic advantage, and help shape how the technology is governed.

We start from strategy. Working with national security veterans who have negotiated directly with America’s allies, partners, and competitors, we study how the AI race reshapes national power and where the real risks lie. The central risk is not competition itself, but misperception: capabilities read as threatening can push others into shortcuts that leave everyone less secure. The task is to advance in ways that build leverage without inviting escalation.

The program also examines what a durable AI governance framework could realistically contain: what can be made transparent, what can be monitored, and what constraints both sides have reason to accept. The goal is not to slow American AI. It is to ensure the United States, not its rivals, shapes that framework.

How the work reaches you.

The analysis takes the forms decision-makers actually use:

  • short policy memoranda for senior leaders,
  • long-form studies grounded in original data and analysis,
  • wargames and simulations that surface problems and pressure-test decisions before they are real,
  • candid, off-the-record roundtables and briefings,
  • and an education program for the policymakers who will make these calls.

Our publications appear in Research as they are released. To receive them, or to request a briefing, get in touch.

Request a briefing

How we approach the work.

Frontier Security Institute (FSI) conducts and supports original studies and analyses at the intersection of frontier AI and U.S. national security, a field that is dynamic, complex, and fast-moving. We believe the best work in this space draws on many viewpoints, and we actively seek out engagement with experts across disciplines and policy communities.

Institutional independence

Each project reflects the work of the team that produced it, and each analysis expresses the views of its author or authors. FSI does not hold a single institutional position. Instead, we provide a platform for rigorous, professional examination of the technical and policy questions that merit the attention of the national security community and the public.

This approach lets FSI’s staff, fellows, and collaborators work from a range of perspectives and in a nonpartisan manner, consistent with our mission.