Executive Summary
This report analyzes how adversarial AI attacks on military AI systems impact strategic stability, identifying vulnerabilities across various military and critical infrastructure AI applications. It highlights that every AI system is vulnerable to attacks that can cause failures, incorrect outputs, or data breaches. The study emphasizes the need for proactive measures and international policy discussions to mitigate these evolving threats.
Why It Matters
This document is crucial for defense and geopolitical analysts as it details how adversarial AI attacks on military systems and critical infrastructure can undermine strategic stability and escalate conflicts. It provides insights into specific vulnerabilities and proposes governance mechanisms to address these emerging threats.
Key Takeaways
- Every AI system is vulnerable to adversarial attacks, which can lead to availability breakdowns, integrity violations, or privacy compromises.
Source evidence · PDF page 4
Adversarial attacks against AI systems involve manipulating data to cause failures of AI systems. These failures could prevent timely access to the AI system, cause wrong and misleading outputs, or disclose confidential data. Every AI system is vulnerable.
- Key military AI systems susceptible to disruption include autonomous weapon systems, military decision-support systems, back-end functions, Nuclear Command, Control and Communications (NC3), and critical national infrastructure.
Source evidence · PDF page 5
Drawing on an expert workshop and a prioritisation exercise, we identified five main types of AI systems whose disruption is likely to impact strategic stability: (1) (Lethal) Autonomous Weapon Systems ([L]AWSs), (2) military decision-support systems, (3) military back-end functions, (4) Nuclear Command, Control and Communications (NC3), and (5) critical national infrastructure.
- Proactive international policy discussions and strengthened systems are necessary to manage the long-term strategic stability risks posed by adversarial AI, even as AI integration into military systems is ongoing.
Source evidence · PDF page 6
Although AI is still being developed and integrated into military systems, meaning adversarial AI may only threaten stability in the longer term, there is a real and current opportunity to proactively strengthen its systems and steer (international) policy discussions:
Strategic Relevance
Adversarial AI poses a significant threat to strategic stability by potentially compromising military AI systems, including autonomous weapons and nuclear command structures. This could shift the offense-defense balance, increase miscalculation risks, and necessitate new governance frameworks to prevent unintended escalation.