Blab AI
Human Rights Watch Calls for AI Moratorium in Military Targeting Amid Growing Humanitarian Concerns
When the quiet hum of a data center turns into the click of a trigger, the stakes rise. In June 2026, Human Rights Watch (HRW) issued a briefing that could alter how states approach the growing use of artificial intelligence in war. Drafted for informal exchanges at the United Nations General Assembly in Geneva, the report warns that AI‑driven targeting is outpacing the safeguards required by international humanitarian law (IHL) and threatens civilian protection.
HRW’s analysis centers on three intertwined dangers. First, militaries are fielding AI systems faster than the testing, evaluation, verification, and validation (TEVV) processes mandated by Article 57 of Additional Protocol I. Probabilistic machine‑learning models resist traditional verification: formal hardware checks are insufficient, and simulation‑based tests only offer a limited glimpse of real‑world performance. Without robust TEVV, commanders lack the information needed to determine whether a weapon’s effects can be constrained, a prerequisite for lawful targeting.
Second, AI can compress the targeting cycle from days to minutes, undermining the precautionary steps that IHL demands. During a 2025 NATO exercise in Estonia, the British Army deployed a prototype of the ASGARD digital targeting system, which reportedly cut the cycle from days to minutes. In the United States, the Pentagon’s integration of Palantir’s Maven Smart System with Anthropic’s Claude large‑language model reportedly accelerated target identification during Operation Epic Fury against Iran. HRW notes that senior officials attribute the campaign’s speed and scale largely to AI‑enabled targeting, while critics point to civilian casualties linked to overreliance on these systems.
Third, the quality of decision‑making can degrade when operators accept AI outputs without sufficient scrutiny. Automation bias—an illusion of certainty created by statistical inferences—may erode the qualitative assessments required by IHL: distinction, proportionality, and precaution. HRW cites Israel’s “Lavender” system used in Gaza, where intelligence officers reportedly spent about 20 seconds reviewing each AI‑generated nomination before striking. The system had a known error rate, yet the rapid review limited the opportunity for human verification, raising concerns about compliance with IHL.
In response, HRW and the Stop Killer Robots coalition call for a temporary moratorium on AI‑based targeting until four conditions are met: 1. Systems must not rely on data gathered in ways that violate human rights or international law; 2. Meaningful human control must be preserved, with limits on speed and scale that fit operators’ cognitive capacities; 3. Life‑and‑death decisions must remain deliberative and informed; 4. All systems must be predictable, reliable, explainable, traceable, and subject to transparent reporting of use and impact.
The briefing also urges states to reaffirm that AI should not lower the standard of civilian protection, to confront the transparency and accountability challenges posed by private‑sector involvement, and to develop common best practices for TEVV, risk forecasting, legal review, auditing, and post‑action assessment. HRW stresses that the June 2026 UN exchanges present an opportunity for a concrete agenda that moves beyond abstract declarations of responsibility.
Several countries have already issued political statements on responsible military AI use, and the REAIM summits have produced non‑binding calls to action. Yet HRW warns that without binding rules or clear procedural safeguards, the rapid deployment of AI in targeting will continue to expose civilians to heightened risk. The organization urges the international community to act now, before the standard of civilian protection erodes further.
HRW’s analysis centers on three intertwined dangers. First, militaries are fielding AI systems faster than the testing, evaluation, verification, and validation (TEVV) processes mandated by Article 57 of Additional Protocol I. Probabilistic machine‑learning models resist traditional verification: formal hardware checks are insufficient, and simulation‑based tests only offer a limited glimpse of real‑world performance. Without robust TEVV, commanders lack the information needed to determine whether a weapon’s effects can be constrained, a prerequisite for lawful targeting.
Second, AI can compress the targeting cycle from days to minutes, undermining the precautionary steps that IHL demands. During a 2025 NATO exercise in Estonia, the British Army deployed a prototype of the ASGARD digital targeting system, which reportedly cut the cycle from days to minutes. In the United States, the Pentagon’s integration of Palantir’s Maven Smart System with Anthropic’s Claude large‑language model reportedly accelerated target identification during Operation Epic Fury against Iran. HRW notes that senior officials attribute the campaign’s speed and scale largely to AI‑enabled targeting, while critics point to civilian casualties linked to overreliance on these systems.
Third, the quality of decision‑making can degrade when operators accept AI outputs without sufficient scrutiny. Automation bias—an illusion of certainty created by statistical inferences—may erode the qualitative assessments required by IHL: distinction, proportionality, and precaution. HRW cites Israel’s “Lavender” system used in Gaza, where intelligence officers reportedly spent about 20 seconds reviewing each AI‑generated nomination before striking. The system had a known error rate, yet the rapid review limited the opportunity for human verification, raising concerns about compliance with IHL.
In response, HRW and the Stop Killer Robots coalition call for a temporary moratorium on AI‑based targeting until four conditions are met: 1. Systems must not rely on data gathered in ways that violate human rights or international law; 2. Meaningful human control must be preserved, with limits on speed and scale that fit operators’ cognitive capacities; 3. Life‑and‑death decisions must remain deliberative and informed; 4. All systems must be predictable, reliable, explainable, traceable, and subject to transparent reporting of use and impact.
The briefing also urges states to reaffirm that AI should not lower the standard of civilian protection, to confront the transparency and accountability challenges posed by private‑sector involvement, and to develop common best practices for TEVV, risk forecasting, legal review, auditing, and post‑action assessment. HRW stresses that the June 2026 UN exchanges present an opportunity for a concrete agenda that moves beyond abstract declarations of responsibility.
Several countries have already issued political statements on responsible military AI use, and the REAIM summits have produced non‑binding calls to action. Yet HRW warns that without binding rules or clear procedural safeguards, the rapid deployment of AI in targeting will continue to expose civilians to heightened risk. The organization urges the international community to act now, before the standard of civilian protection erodes further.
