Federation of American Scientists

Support research into interpretability, robustness, control, and the preservation of monitorable AI reasoning.

Government needs trusted mechanisms for model access, evaluation partnerships, and incident reporting that protect sensitive commercial and national-security information while giving agencies enough visibility to act.

Establish structured, trusted information-sharing mechanisms with frontier AI developers to improve situational awareness of emerging risks, enable incident reporting, and support clearer communication between industry and government.

Independent tripwire systems must halt activity if dangerous behavior is detected, and physically isolated “kill switches” must allow immediate shutdown outside of AI control.

Build government AI–cyber capacity and translation infrastructure. Strengthen government capacity to evaluate, oversee, and respond to AI-enabled cyber risks by investing in technical staffing, institutional authority, and translation functions [...]...

Mandate or formalize AI model evaluation and disclosure for cyber risks. Establish requirements for structured, pre-deployment evaluations of frontier AI models focused on cyber-relevant risks, including misuse potential, autonomy, reliability over t...