Carla B. Abdo-Katsipis

Artificial intelligence is transforming warfare, but not because it is replacing human judgment. Its most consequential effect is the compression of decision time.

Public debate about military AI has largely focused on autonomous weapons and the prospect that machines may one day decide whom to kill without meaningful human intervention. Diplomats have debated restrictions on so-called “killer robots,” while ethicists have questioned whether lethal decision-making should ever be delegated to algorithms. These concerns are important, but they risk obscuring a transformation that is already reshaping armed conflict. The defining military impact of artificial intelligence is not autonomy. It is speed.

For centuries, military operations were constrained by the pace at which information could be collected, interpreted, and verified. Reconnaissance patrols required time to report their observations. Aerial imagery demanded painstaking analysis. Communications intercepts had to be translated, contextualized, and corroborated against other intelligence before commanders could determine whether military action was justified. Intelligence itself functioned as an institutional brake on the use of force because human beings could process only a finite amount of information before making operational decisions.

Artificial intelligence is steadily eroding that constraint.

Modern decision-support systems can integrate satellite imagery, drone surveillance, electronic intercepts, geospatial intelligence, communications metadata, and historical operational data into a single analytical process capable of identifying relationships across enormous datasets in minutes rather than days.

These systems do not necessarily decide whom to attack. Rather, they assist analysts through the use of suggestion models, organizing information, identifying patterns, highlighting anomalies, and generating candidate military objectives against a pre-existing library of defined targets. Their principal contribution is not autonomous targeting but a dramatic expansion in the speed and scale of intelligence analysis.

This acceleration offers clear military advantages. Commanders can develop a more comprehensive understanding of the battlefield, respond more rapidly to emerging threats, and exploit fleeting operational opportunities. Yet speed also changes the character of military decision-making itself. If technology enables intelligence organizations to generate hundreds of potential military objectives where they previously generated dozens, the limiting factor in warfare is no longer intelligence collection. It becomes the institutional capacity to evaluate those recommendations responsibly before force is employed.

This shift has profound implications for civilian protection. Essentially, the pre-existing library of information that defines targets requires precision and subtlety—or civilians can be suggested as targets, too.

International humanitarian law was intentionally designed to regulate armed conflict regardless of technological change. The principles of distinction, proportionality, and precaution remain fully applicable whether intelligence is gathered through reconnaissance patrols, satellites, or AI-supported decision-support systems. Commanders remain legally obligated to distinguish civilians from combatants, verify that proposed objectives constitute lawful military targets, assess whether anticipated civilian harm would be excessive in relation to the expected military advantage, and take all feasible precautions before authorizing an attack. Artificial intelligence changes none of these obligations. It changes the conditions under which they must be fulfilled.

The conflicts in Gaza and Lebanon provide an important lens through which to examine this transformation. They are not significant because they are unique, but because they represent the first widely documented examples of AI-supported intelligence being employed during sustained, high-intensity combat. Investigative reporting has described the reported use of AI-assisted decision-support systems—including platforms publicly identified as Gospel (Habsora) and Lavender—to assist intelligence analysts in identifying and prioritizing potential military objectives. Israeli officials have acknowledged integrating artificial intelligence into aspects of intelligence analysis while disputing or declining to confirm many operational details described in public reporting. Regardless of the precise capabilities of any individual system, the broader trend is unmistakable: AI-assisted decision support has become an increasingly important feature of contemporary military operations.

The significance of Gaza and Lebanon extends beyond the technologies reportedly employed. Together, they illustrate two fundamentally different intelligence problems.

Gaza confronted Israeli planners with one of the most complex urban battlespaces in modern warfare. Hamas operated within an exceptionally dense civilian environment in which residential neighborhoods, hospitals, schools, humanitarian facilities, commercial infrastructure, and an extensive tunnel network frequently existed in close proximity to military assets. Intelligence analysts were required to distinguish lawful military objectives from civilian objects while battlefield conditions changed continuously and large numbers of civilians were displaced. Under such conditions, the ability to process surveillance imagery, communications intercepts, and geospatial intelligence rapidly offered obvious operational advantages.

Lebanon presented a different challenge. Hezbollah’s military infrastructure is geographically dispersed across southern Lebanon, where launch sites, logistics networks, command facilities, and communications systems are distributed across villages, transportation corridors, agricultural terrain, and mountainous landscapes. Rather than navigating extreme urban density, intelligence organizations faced the problem of identifying concealed military networks embedded within a broad geographic area. Here, AI-assisted intelligence offered value by identifying relationships across dispersed datasets that might otherwise require extensive human effort to recognize.

Although these operational environments differ, they reveal the same institutional challenge.

Artificial intelligence does not eliminate uncertainty.

It relocates it.

The uncertainty no longer lies primarily in whether militaries can collect enough information. It lies in whether military institutions can evaluate an expanding volume of information carefully enough to satisfy their legal and ethical obligations before operational tempo overtakes human judgment.

This distinction is the central argument of this article.

Much of the debate surrounding military AI continues to ask whether machines will eventually replace human decision-makers. The more urgent question is whether human decision-makers will continue to have sufficient time to exercise meaningful judgment at all.

Time as a Legal Safeguard

The significance of artificial intelligence extends beyond the battlefield. It challenges a fundamental assumption underlying the modern law of armed conflict: that military institutions possess sufficient time to exercise informed legal judgment before force is used.

International humanitarian law was deliberately drafted to be technology-neutral. Whether intelligence is gathered through reconnaissance patrols, satellites, or AI-assisted decision-support systems, the same legal obligations apply. Commanders remain responsible for distinguishing civilians from combatants, verifying that proposed objectives constitute lawful military targets, ensuring that anticipated civilian harm is not excessive in relation to the concrete and direct military advantage anticipated, and taking all feasible precautions before an attack is authorized.[i]

Artificial intelligence changes none of these legal standards.

What it changes is the environment in which they must be applied.

For much of modern military history, intelligence analysis imposed a natural limit on operational tempo. Information had to be collected, compared, debated, and verified before commanders could decide whether military action was justified. Analysts questioned contradictory reporting, intelligence officers assessed confidence levels, and legal advisers evaluated whether available evidence satisfied the requirements of international humanitarian law. The process was often slow, but that slowness served an important institutional function. It created opportunities to identify errors, reconsider assumptions, and delay military action until commanders possessed greater confidence in the available intelligence.

Artificial intelligence fundamentally alters this relationship.

Machine-learning systems can process enormous quantities of information almost instantaneously, correlating satellite imagery, drone surveillance, electronic intercepts, communications metadata, and historical intelligence into recommendations for human review. As a result, intelligence organizations are capable of generating significantly more candidate military objectives than would have been possible through traditional analytical methods. This increased analytical capacity offers genuine operational advantages, particularly in rapidly evolving conflicts where opportunities may disappear within minutes.

Yet increased speed also creates new institutional pressures.

As commanders and legal advisers confront larger volumes of intelligence under compressed timelines, preserving the quality of legal review becomes increasingly difficult. The challenge is no longer simply identifying potential military objectives. It is ensuring that each recommendation receives the degree of scrutiny required before force is employed.

The International Committee of the Red Cross has recognized this emerging tension. In its analysis of AI-assisted military decision-making, the ICRC argues that accelerating operational tempo can increase the risks of miscalculation, escalation, and civilian harm if institutions do not preserve adequate time for human deliberation. It therefore recommends maintaining what military planners have traditionally described as “tactical patience”—deliberately slowing elements of operational planning when necessary to improve situational awareness, consider alternative courses of action, and reduce risks to civilians.[ii]

This recommendation deserves greater attention than it has received.

Much of the public discussion surrounding military AI assumes that technological progress necessarily requires ever-faster operational decisions. The ICRC’s analysis suggests the opposite. In some circumstances, responsible military decision-making may require institutions to deliberately preserve moments of analytical pause even as technological capabilities continue to accelerate.

Seen in this light, the central challenge posed by artificial intelligence is not whether machines will eventually replace human decision-makers. It is whether military organizations can preserve meaningful human judgment when the pace of warfare increasingly exceeds the pace at which institutions were designed to operate.

Civilian Protection Begins Before the Strike

Public discussions of civilian casualties often begin after an attack has occurred. Investigators assess casualty figures, humanitarian organizations document damage, and legal experts debate whether a particular strike complied with international humanitarian law.

These assessments are indispensable.

They are also retrospective.

By the time civilian harm can be measured, the most consequential decisions have already been made.

Civilian protection begins during intelligence collection.

Before any weapon is launched, military institutions must determine whether available intelligence is sufficiently reliable, whether the proposed objective makes an effective contribution to military action, whether civilians are likely to be present, and whether additional information should be obtained before force is employed. These decisions determine not only whether an attack proceeds but also how risk is distributed between military necessity and civilian protection.

Artificial intelligence increasingly shapes this earlier stage of military planning. Decision-support systems organize information, identify patterns across multiple intelligence sources, generate candidate targets, and assist analysts in prioritizing competing operational requirements. Used appropriately, these capabilities can improve situational awareness and enable commanders to make more informed decisions. Used without adequate institutional safeguards, however, they may also increase pressure to convert analytical recommendations into operational decisions before uncertainty has been sufficiently resolved.

The concern is therefore not that AI inevitably produces incorrect intelligence.

Nor is there persuasive public evidence that AI systems independently determine whether military action complies with international humanitarian law.

Rather, the concern is institutional.

As target generation accelerates, military organizations may struggle to preserve the rigorous verification processes upon which civilian protection depends. Intelligence gaps, outdated information, contradictory reporting, or uncertainty that might once have prompted additional review may instead be carried forward under pressure to maintain operational momentum. The danger lies less in algorithmic failure than in the possibility that institutional safeguards become compressed alongside the targeting cycle. The experiences of Gaza and Lebanon illustrate this challenge in different operational environments. Gaza demonstrates the extraordinary complexity of conducting military operations in one of the world’s most densely populated urban areas, where military objectives and civilian life frequently overlap. Lebanon illustrates the equally demanding task of identifying dispersed military infrastructure concealed within civilian communities and difficult terrain. Although the intelligence problems differ, both conflicts reveal the same reality: artificial intelligence expands the capacity to process information, but it cannot eliminate uncertainty or replace the legal and ethical judgment required before force is used.

---

[i] Protocol Additional to the Geneva Conventions of 12 August 1949 (Protocol I), arts. 48, 51(5)(b), and 57.

[ii] International Committee of the Red Cross, International Humanitarian Law and the Challenges of Contemporary Armed Conflicts, in particular the section “Preserving Time and Space for Human Deliberation,” which discusses AI decision-support systems, the risks of accelerated military planning, and the importance of “tactical patience.”