Risks of Adversarial Attacks on Military Artificial Intelligence Systems

A research study prepared by RAND Europe and funded by the UK Foreign, Commonwealth & Development Office examined how adversarial attacks on military artificial intelligence systems could threaten strategic stability and nuclear and conventional command and control systems, offering various recommendations to governments.

Military Artificial Intelligence and Strategic Stability

While strategic competition in military artificial intelligence is intensifying rapidly, global governance fails to keep pace. As reliance on AI-driven decision-making processes increases, attacks that disrupt or manipulate military and critical infrastructure systems undermine strategic stability.

Credibility and Escalation Impacts of Attacks

The perception that AI systems are vulnerable to adversarial attacks weakens defense credibility among allies and adversaries alike. Such interventions tilt the offense-defense balance in favor of the offense, increasing risks of miscalculation and mistrust.

Key Policy Recommendations for Governments

Governments must address adversarial AI threats not merely as a cybersecurity or assurance issue, but directly as a strategic stability risk. It is vital to evaluate these risks within national governance mechanisms and international platforms.

Critical Infrastructure and Nuclear Security Measures

It is vital to strengthen nuclear and dual-use command, control, and communication infrastructures through robust design, resilience, recovery planning, and fallback options. Clear red lines must be established with rivals, and communication channels must be kept open.