AI Emergency Preparedness: Examining the federal government's ability to detect and respond to AI-related national security threats

TL;DR

The paper addresses the challenge of detecting and responding to AI-related national security threats. It adopts a scenario-based framework to explore three plausible risk pathways: loss of control from AI-accelerated R&D, cybersecurity-enabled weight theft, and biological weaponization from open model weights. For each scenario, it assesses detection, prevention, and response capabilities, identifies systemic gaps within the US National Preparedness System, and offers concrete measures such as frontier developer interviews, AI emergency hotlines, embedded auditors, a dedicated AI emergency preparedness unit, and strengthened international coordination. The work highlights that proactive emergency preparedness can improve risk understanding, accelerate detection and escalation, and enable timely crisis response while preserving innovation.

Abstract

We examine how the federal government can enhance its AI emergency preparedness: the ability to detect and prepare for time-sensitive national security threats relating to AI. Emergency preparedness can improve the government's ability to monitor and predict AI progress, identify national security threats, and prepare effective response plans for plausible threats and worst-case scenarios. Our approach draws from fields in which experts prepare for threats despite uncertainty about their exact nature or timing (e.g., counterterrorism, cybersecurity, pandemic preparedness). We focus on three plausible risk scenarios: (1) loss of control (threats from a powerful AI system that becomes capable of escaping human control), (2) cybersecurity threats from malicious actors (threats from a foreign actor that steals the model weights of a powerful AI system), and (3) biological weapons proliferation (threats from users identifying a way to circumvent the safeguards of a publicly-released model in order to develop biological weapons.) We evaluate the federal government's ability to detect, prevent, and respond to these threats. Then, we highlight potential gaps and offer recommendations to improve emergency preparedness. We conclude by describing how future work on AI emergency preparedness can be applied to improve policymakers' understanding of risk scenarios, identify gaps in detection capabilities, and form preparedness plans to improve the effectiveness of federal responses to AI-related national security threats.