Human Rights Watch Calls for Moratorium on AI Targeting Systems

Military AI deployment outpaces safety evaluation

Militaries worldwide are integrating artificial intelligence into life-and-death decisions—target selection, force calculations, civilian harm assessments—faster than they can verify these systems work safely or legally, according to a briefing Human Rights Watch prepared for United Nations discussions in Geneva.

The organization, which co-founded the Stop Killer Robots campaign, is calling for an immediate moratorium on using any AI system as the basis for targeting decisions until governments establish testing standards, transparency requirements, and accountability mechanisms.

The briefing, prepared for informal UN exchanges held in June 2026, draws on Human Rights Watch's decade of research into autonomous weapons and its recent documentation of Israel's automated targeting tools in Gaza. The organization identified three urgent concerns that existing international humanitarian law may not adequately address.

Why it matters

The battlefield has become a de facto testing ground for AI systems that resist traditional evaluation methods. Unlike conventional weapons, machine learning models can only be partially assessed through simulations, and their opacity limits reliable evaluation. When militaries field systems without understanding their behavior, they cannot ensure compliance with laws designed to protect civilians. The standard of civilian protection risks drifting downward as probabilistic outputs replace the qualitative legal judgments international law requires.

Three core problems

First, testing and evaluation cannot keep pace with adoption. AI systems behave unpredictably, and formal testing methods fall short. Human Rights Watch noted that parties to conflicts are deploying capabilities before understanding how they will perform in actual use.

Second, AI enables attacks at speeds and scales that make meaningful precautions difficult or impossible. When systems compress decision timelines, they undermine the deliberative assessment international humanitarian law mandates. A party cannot excuse failures to verify targets or assess civilian harm by citing the speed of its own technology.

Third, AI degrades rather than augments human judgment. Automation bias, machine learning opacity, and the substitution of statistical inference for legal reasoning all erode decision-makers' capacity to apply the law. AI systems can generate false information and amplify discriminatory patterns in ways that are hard to detect.

Documented use cases

The briefing cited specific examples: Israel's Lavender and Gospel systems for target classification, Ukraine's Saker Scout and Russia's Zala Lancet autonomous munitions, and the U.S. military's use of AI to identify targets and accelerate decisions in operations involving Iran. Human Rights Watch has separately documented four digital tools Israel used in Gaza that rely on faulty data and inexact approximations, increasing civilian harm risks.

What states should do

Beyond the moratorium, Human Rights Watch recommends states reaffirm that AI adoption must not lower civilian protection standards. Governments should establish common practices for testing, verification, legal reviews, and post-action assessments. They should acknowledge that AI introduces novel transparency and accountability challenges due to technical design, licensing arrangements, and private sector involvement.

Future discussions should focus on applying international law to AI in force decisions, establishing red lines for AI use in conflict, and prioritizing applications of greatest humanitarian concern. States should also create accessible mechanisms for remedy and redress.

The briefing emphasized that "AI in the military domain" encompasses heterogeneous applications raising distinct concerns—from computer vision for target identification to large language models summarizing intelligence data to drone swarm coordination. Each use case demands specific policy responses.

The details were first reported by Human Rights Watch in a background briefing prepared for the June 2026 UN informal exchanges on artificial intelligence in the military domain.