AI in warfare: why military decision-making is the most critical risk
The rapid integration of artificial intelligence into military command structures is reshaping how wars are planned, launched, and conducted, but experts warn that the most critical danger lies in delegating key decisions to machines that do not understand escalation, proportionality, or de-escalation the way humans do. AI-driven decision support systems can process vast volumes of satellite imagery, sensor data, and communications far faster than human staffs, yet their opacity and error propagation risks mean that a single flawed recommendation can cascade through several automated steps and push a crisis toward unintended conflict or even nuclear confrontation.
In current conflicts, militaries already use AI to fuse data from drones, satellites, radars, and electronic intelligence in order to detect threats and generate targeting options at unprecedented speed, which allows commanders to engage opponents more quickly but also compresses the time available to assess the situation and weigh alternatives that might spare civilians or avoid escalation. Military planners present these systems as tools to improve situational awareness and shorten decision cycles, yet humanitarian organizations such as the International Committee of the Red Cross argue that this acceleration can undermine the deliberate planning processes that are designed to protect civilians and ensure compliance with international humanitarian law.
Researchers who tested advanced language-model-based agents in simulated crises found that these systems tend to choose escalation in high-risk diplomatic or military scenarios, including the use of nuclear weapons even when no clear trigger exists, which reinforces concerns that AI does not naturally favor restraint and could push states toward more aggressive postures if its outputs are trusted too readily. This risk is magnified by the difficulty, and sometimes impossibility, of tracing how complex models reach specific recommendations, making it hard for commanders to verify whether an AI-generated option respects legal obligations, operational realities, and political constraints before acting on it.
Ethical and legal debates increasingly focus on the need to keep humans “in the loop” or at least “on the loop” for all decisions involving the use of force, yet practice is drifting toward greater automation, as illustrated by US programs such as Maven, where AI systems already propose targets and allocate the most suitable weapon platforms before a human gives final approval. Critics fear that, under time pressure or in complex battlespaces, human operators may simply validate AI recommendations rather than exercise independent judgment, turning formal oversight into a rubber stamp while responsibility for mistakes remains unclear if civilian casualties or disproportionate attacks occur.
Humanitarian actors emphasize that traditional military planning doctrines, which encourage commanders to take time to study context, adversary behavior, and civilian presence, cannot be outsourced to algorithms because the reflective process itself helps identify options that de-escalate or avoid the use of force. Slowing decision cycles, even when AI tools are available, can give forces room to reposition, resupply, or choose non-violent courses of action, whereas a relentless drive for speed risks locking both sides into rapid-fire exchanges with little space for diplomacy or reassessment.
At the same time, some experts argue that carefully designed AI decision-support tools could help protect civilians by processing large datasets about population movements, protected sites, or previous patterns of harm, provided that their outputs remain transparent, contestable, and subordinate to human judgment. Global investment in military AI, however, is soaring in the United States, China, and Europe, which increases pressure to deploy systems quickly and may outpace efforts by organizations such as the ICRC and advocacy groups like “Stop Killer Robots” to secure binding international rules that enshrine meaningful human control over targeting and escalation decisions.
Several analysts warn that less technologically advanced states or emerging powers might see AI-enhanced weapons and command tools as a way to offset conventional military disadvantages, making autonomous or semi-autonomous systems attractive in regional rivalries and thus raising the risk of miscalculation in already fragile security environments. In this context, the central question is not only whether to use AI in war, but how to design governance frameworks, transparency requirements, and operational safeguards that ensure algorithms augment human judgment instead of silently redefining who decides when to threaten, escalate, or employ lethal force.
-
18:25
-
18:10
-
17:47
-
17:30
-
17:24
-
17:15
-
17:15
-
17:14
-
17:11
-
17:01
-
17:00
-
17:00
-
16:52
-
16:49
-
16:45
-
16:30
-
16:30
-
16:22
-
16:15
-
16:13
-
16:00
-
15:59
-
15:45
-
15:36
-
15:30
-
15:15
-
15:14
-
14:58
-
14:41
-
14:39
-
14:20
-
14:15
-
14:05
-
14:01
-
13:45
-
13:30
-
13:15
-
12:58
-
12:41
-
12:24
-
12:15
-
11:52
-
11:47
-
11:25
-
11:18
-
11:15
-
11:11
-
11:01
-
10:47
-
10:42
-
10:37
-
10:32
-
10:20
-
10:15
-
10:00
-
10:00
-
09:43
-
09:42
-
09:25
-
09:18
-
09:09
-
08:51
-
08:47
-
08:41
-
08:30
-
08:15
-
21:00
-
19:00
-
18:45
