AI-on-AI Perfidy and the Law of Armed Conflict

https://lieber.westpoint.edu/ai-on-ai-perfidy-law-armed-conflict/

Publish Date: 2026-06-08 10:38:00

Source Domain: lieber.westpoint.edu

Scenario of AI-Enabled Deception: An AI surveillance system identifies a convoy as medical, leading to a recommendation for non-engagement, which is later exploited by the convoy to deploy devastating loitering munitions.
Legal Framework and Perfidy: The article explores the legal definition of perfidy under Article 37 of 1977 Additional Protocol I, which prohibits deceptions that exploit protected statuses. Traditional perfidy involves human deception, but the article asks whether machine-mediated deceptions fit this framework.
AI-to-AI Perfidy Definition: The term “AI-to-AI perfidy” describes the intentional simulation of protected status by one AI system to deceive another, resulting in restrained actions that the deceiver then exploits to cause harm.
Functional Understanding of Reliance: The article argues for a functional approach to reliance, where the key focus is on whether an AI or human target acts in accordance with legal protection signals, maintaining that both human and machine reliance are relevant for the perfidy prohibition.
New Features of Machine-Mediated Deception: The scalability of AI systems and their ability to manipulate algorithmic recognition processes bring new challenges to the concept of perfidy, including exploiting statistical models and creating deceptive signals undetectable by human observers.
Challenges and Doctrine Gap: While deliberate AI-to-AI perfidy fits within existing legal prohibitions, new challenges arise in proving intent, preventing unforeseen exploitation, and distinguishing it from other forms of electronic warfare.
Avoiding a Doctrinal Vacuum: Ensuring that the prohibition against perfidy encompasses AI-mediated deceptions is crucial to maintaining the integrity of protected signals in warfare and avoiding legal gaps that could undermine the law of armed conflict.