A decision used to take days in the interim between obtaining intelligence and carrying out a strike. It supposedly takes seconds now. The US military’s position in 2026 is defined more by that compression than by any one weapon system or change in policy.
Over 1,000 targets inside Iran were hit during the first 24 hours of Operation Epic Fury. That figure had risen above 13,000 by April 9th. The Maven Smart System, a Palantir-built platform that grew out of a 2017 Pentagon initiative and is subtly positioned as the most significant targeting tool the US military has ever used at scale, is responsible for that volume.
The system performs tasks that used to take hours for human analysts. It gathers data from radar feeds, satellite imagery, drone footage, and signals intelligence, applies machine-learning models to all of it, gives detected targets confidence scores, and then suggests preferred weapons and strike options. After reviewing the output, a human officer either approves the shot or moves it up the chain. A person still makes decisions on paper. In reality, it’s worthwhile to consider what that decision actually looks like when you’re processing one every 3.6 seconds.
Brad Cooper, who oversaw US Central Command during the Iran campaign, attested to this when he said that procedures that “used to take hours and sometimes days” now happen in a matter of seconds. In simpler terms, Craig Jones, a kill chain analyst, suggested that the system now functions “much quicker in some ways than the speed of thought.” More attention should be paid to that phrase than it usually receives.
Beyond its political and strategic aspects, the Iran conflict is particularly noteworthy because it serves as the first real large-scale field test of a military apparatus with AI integration. A partial preview was provided by Ukraine, where the Maven system was tested to improve battlefield visibility for Ukrainian commanders. However, the system’s more sophisticated targeting features were purposefully removed. With Israel using its own AI systems to create targeting databases at a scale and speed that unnerved outside observers, Gaza also provided early indications of where this was going.
Many of those limitations were lifted during the Iran operation. And the outcomes have compelled a challenging discussion.
Human rights monitoring organizations have documented over 1,700 civilian deaths. 175 people were killed in the deadliest single incident, a strike on a Minab primary school, the majority of whom were girls between the ages of seven and twelve. According to a preliminary investigation reported by the New York Times, satellite imagery had previously revealed that the school shared a compound with a military installation; however, this was no longer the case in 2016. The school’s website was up and running. Google Maps showed it. How those details did not register prior to the strike’s authorization is still unknown.
It’s unclear if the Maven system had a direct influence on that choice. However, the more general issue it brings up is structural. By definition, a platform built to generate 1,000 targeting decisions per hour is not intended to slow down and investigate the origin of a satellite image taken ten years ago. In many respects, speed and scrutiny are at odds with one another.
The human review process runs the risk of becoming a formality rather than a true check when meeting a decision quota becomes the operational goal. Before authorizing strikes, an Israeli intelligence officer using a comparable AI targeting system reported reviewing each file in less than a minute. This wasn’t because the system was perfect, but rather because the speed required it.
Automation bias is the well-documented tendency for human operators to follow machine recommendations rather than challenge them, particularly when under time pressure. This phenomenon has been identified by international lawyers and conflict researchers.
The Pentagon has maintained an unrepentant stance, at least in public. The phrase “maximum lethality, not tepid legality” was used by Secretary of War Pete Hegseth to describe the Iran campaign; it has since become somewhat of a doctrine marker. It makes it very evident where institutional priorities are. Rewriting the rules of engagement may not always prioritize accountability.
It’s important to acknowledge that there is a valid counterargument. Advocates of AI in the military community contend that by increasing precision, faster processing of better targeting data can actually lessen harm to civilians. A well-calibrated machine is not intrinsically safer than a human analyst going through a backlog of imagery at two in the morning. Whether or not humans are always superior is not the question. It concerns whether the current system is dependable enough to be trusted at this scale, given its current speed and oversight mechanisms.
That’s still an open question. It has been loudly raised by the Iran campaign. And the solution will be a major challenge for any further advancements in military AI.
