Our work is written for the rooms where frontier capability becomes doctrine, acquisition, and operational decision — and in the formats those rooms can use.
Two communities — the frontier AI labs and the National Security Enterprise — are converging on a problem neither was built for. Frontier Security Institute does the work in between: aggregating findings across labs that cannot share with one another directly, stress-testing capability claims against the missions that will use them, and translating the result into the forms doctrine, acquisition, and operations require.
We are part of the Center for AI Safety, which funds our work. AI safety is the larger mission. Frontier Security Institute carries that mission into the rooms where the National Security Enterprise makes its decisions.
A condition without modern precedent
"AI, as a national security technology, now lives entirely inside the commercial sector — a condition without modern precedent."
If no one clears the institutional space between frontier labs and federal decision-makers, the gap widens with every model release. The cost: doctrine that does not catch up, acquisition pathways that do not open, strategic surprises that did not need to be surprises, and the opportunity for adversaries to co-opt American innovation.
Senior leadership from the Uniformed Services, Cabinet departments, national security research, and the Intelligence Community. Our principals have made the calls our partners are now making.
An independent nonprofit. Objective findings are the product. No commercial pressure, no political alignment, no contracting incentive.
Fluent in frontier AI research practice and in the operational tempo, language, and constraints of the National Security Enterprise. The translation does not lose technical fidelity — and does not stall on jargon.
DC-based, with decades of insider experience across the Pentagon, the Intelligence Community, Commerce, Congress, and allied institutions. Fluent in the tempo, language, and constraints of national security.
If you build, deploy, govern, or oversee AI in a national security context, that is what we are built for.