Military AI Should Be Judged on Compliance, Not Just Risk

As artificial intelligence becomes embedded in military operations—from intelligence analysis to target verification—the legal debate has focused overwhelmingly on the risks these systems pose to civilian protection. But a new analysis argues this approach captures only half the picture.

Military AI-enabled decision-support systems (AI-DSS) should be evaluated not only by their potential to undermine compliance with international humanitarian law (IHL), but also by their capacity to help commanders fulfill existing legal obligations to protect civilians, according to research from West Point's Lieber Institute.

Why it matters

This reframing has practical consequences. If an AI system materially improves target verification or reduces false positives without excessive military cost, its use may become legally necessary under IHL's requirement to take "all feasible precautions." The question shifts from whether AI is risky to whether it enables better compliance than unaided human decision-making under the same operational pressures.

Precaution as the Legal Standard

Article 57 of Additional Protocol I requires commanders to take all feasible precautions throughout military operations—verifying targets are lawful military objectives, choosing means and methods that minimize civilian harm, and canceling attacks when circumstances change. Compliance depends not just on final decisions but on the information-gathering and assessment processes leading to them.

Recent scholarship by Renato Wolf demonstrates that AI-DSS affect precautionary obligations throughout target verification, not just at the moment of human approval. Jessica Dorsey and Luke Moffett note that AI is becoming embedded across the entire targeting process. This means legal assessment cannot focus solely on algorithmic recommendations at decision points—it must examine how systems operate throughout the chain.

Crucially, what constitutes a "feasible" precaution depends on available technology. An AI capability that materially improves target verification may alter what precautions commanders are legally required to take.

The Right Comparison

Current debates often measure AI against idealized human performance. But IHL requires precautions "in the circumstances ruling at the time"—meaning conditions of uncertainty, incomplete information, time pressure, and stress.

The relevant legal comparison is therefore between AI-assisted and unaided human decision-making under identical operational conditions. The question is not whether AI achieves perfect accuracy, but whether it enables more effective implementation of precautionary obligations than would otherwise be possible.

This matters because AI-DSS are designed to support, not replace, human judgment. By integrating intelligence from multiple sources and analyzing volumes of data beyond human capacity, these systems may improve target verification, identify civilian presence, and help commanders respond when attacks must be modified or canceled.

Beyond Individual Attacks

The analysis also raises a broader question: Should AI's humanitarian implications be assessed at the campaign level, not just attack-by-attack?

AI may reduce civilian risk per attack while simultaneously enabling more attacks over the same period. The aggregate humanitarian outcome becomes ambiguous—improved precision per engagement could be offset by increased operational tempo. Conversely, greater efficiency might shorten conflicts and reduce overall civilian exposure to violence.

Which effect predominates remains an open empirical question. But the operational dimension suggests that assessing AI exclusively through individual attacks provides only a partial account of its humanitarian implications.

The research emphasizes that this reframing does not diminish concerns about accountability, meaningful human control, or algorithmic reliability. Rather, it places those risks within the broader legal framework that IHL actually provides—one that requires affirmative measures to protect civilians, not just prohibitions on harmful conduct.

These details were first reported by the Lieber Institute for Law and Land Warfare at West Point.