Stress-testing
Independent capability evaluation
Rigorous, non-commercial assessment of frontier model capabilities in scenarios relevant to national security decision-makers.
Read more →For frontier labs and AI safety researchers: the channel from technical findings to the doctrine, acquisition, and operational decisions that determine how AI gets fielded — or whether it does.
Three modes of engagement with frontier AI labs and the broader research community.
Stress-testing
Rigorous, non-commercial assessment of frontier model capabilities in scenarios relevant to national security decision-makers.
Read more →Translation
Turning research outputs into the formats operators, policymakers, and acquisition leaders can act on without losing technical fidelity.
Read more →Convening
A neutral institutional venue for direct conversation between frontier labs and the national security enterprise — with editorial standards on both sides.
Read more →Frontier Security Institute is an independently governed affiliate of the Center for AI Safety. The parent's mission — reducing societal-scale risks from artificial intelligence — is the field we extend into the National Security Enterprise.
Visit the Center for AI SafetyEvaluations, red-teams, joint convenings, technical advisory — if your work shapes what frontier AI can do, ours determines how it gets used.