00:09In early 2026, the United States military struck roughly a thousand targets in the first 24 hours
00:16of its operations against Iran. A thousand in a single day. For context, during the Gulf War in
00:251991, hitting that many targets would have taken weeks of planning and thousands of personnel
00:30working around the clock. During the Iraq War in 2003, the Combined Air Operations Center needed
00:37over a thousand staff just to coordinate a few hundred strikes. But in 2026, the military did it
00:45with AI. A system called the MAVEN Smart System scanned satellite imagery, drone footage, radar
00:54data and intelligence reports, then built targeting packages faster than any room of analysts ever
01:01could. And just weeks before that, the Under Secretary of the Army had stood at a podium and
01:08announced something that would have sounded impossible five years earlier.
01:32The Army's AI had enabled a division to prosecute 15 different targets in a single hour. To understand
01:40how we got here, you have to understand something that never makes it into the headlines. The hardest
01:47part of modern warfare isn't pulling the trigger. It's deciding what to shoot, when to shoot it, and
01:54with what. The military calls this process the kill chain. Six steps. Find the target. Fix its exact
02:03location. Track its movement. Decide whether it's a valid target. Choose the right weapon. Fire. Then assess
02:12whether you actually hit it. Every one of those steps used to require different people, different
02:18systems, different buildings, sometimes different countries. In World War II, the whole cycle from
02:25aerial reconnaissance photos to assembled strike packages could take weeks. A plane would fly over
02:32enemy territory, take photographs on actual film, fly back, land, develop the film, hand it to analysts who
02:41would study it for days, then pass their findings up the chain of command.
03:07By the time bombs dropped, the targets had often moved, or the situation had changed entirely. The Gulf War
03:15compressed that timeline from weeks to hours. Better satellites, faster communications, precision
03:22guided munitions. But even then, the US struggled. Saddam Hussein's mobile scud launchers would fire a
03:31missile and relocate within minutes. By the time American forces identified the launch site, the launcher
03:38was already gone. The technology to find targets was getting faster, but the humans processing the
03:44information couldn't keep up. By the Iraq War in 2003, things sped up again. On April 7th, Saddam Hussein was
03:55reportedly spotted in a Baghdad neighborhood. Forty-eight minutes later, a B-1 bomber had already cratered the
04:03location. That was considered extraordinary at the time. But it still required rooms full of analysts
04:10tests, staring at screens, passing sticky notes, toggling between disconnected systems, making phone calls
04:19to confirm targets. It worked, but it didn't scale. That's the problem AI was built to solve.
04:46In April 2017, the Pentagon launched Project Maven. The original idea was straightforward.
04:55The military had more drone footage than its analysts could ever watch. Thousands of hours pouring in
05:02daily from surveillance flights over Iraq, Syria, and East Africa. Maven used computer vision, a type of AI
05:11designed to recognize objects in images and video, to scan that footage automatically. It could spot a
05:18tank, a truck, a radar installation, and flag it for a human analyst to review. What used to take a
05:27team of
05:27people hours to sift through, Maven could process in seconds. But Maven didn't stay small. Google was the
05:35original tech partner, and when employees found out their work was being used for military targeting,
05:41they revolted. Google pulled out in 2018, but Palantir, the defense technology company,
05:50had no such hesitation. They took over and built Maven into something far bigger. A full command and
05:57control platform that pulls data from over 150 sources. Satellite imagery, drone video, radar, infrared
06:05sensors, signals intelligence, geolocation data, all of it feeding into one system that can identify
06:13targets, prioritize them by importance, recommend the right weapon for each one, and even communicate
06:21directly with firing units in the field.
06:44The 18th Airborne Corps at Fort Bragg became Maven's primary testing ground through a series of
06:50exercises called Scarlet Dragon. The first live fire test in December 2020 was the proof of concept.
06:59AI identified a decommissioned tank in satellite images, a human confirmed the selection, and the
07:06system sent a fire order to an M142 HIMARS. It worked, but it took 743 minutes. Over 12 hours to
07:16kill one
07:17target that wasn't shooting back. With each new exercise, the system got faster. The AI learned. The
07:24operators learned how to work with it. And the results were staggering. A Georgetown University
07:30investigation found that Maven enabled the same targeting output that previously required close to
07:362000 personnel with just 20 human operators. The same work. Roughly one percent of the people.
07:44If breakdowns like this are your thing, hit subscribe. We cover this stuff regularly. But here's the part
07:52that should concern everyone, including the people building these systems. Maven doesn't just make the
08:00process faster for your side. It changes what's possible in combat. The system can nominate up to a
08:07thousand potential targets per hour, each one paired with a recommended weapon, platform availability,
08:14flight time calculations, and the locations of friendly forces to prevent accidental strikes on your own
08:20troops.
08:41A human still has to approve each target. That's the official rule. But when the machine is suggesting a
08:48thousand targets an hour and the pressure of combat demands rapid decisions, the difference between
08:55human approval and human rubber stamping gets razor thin. And 2026 proved exactly that.
09:05When the U.S. hit those thousand targets in the first day of operations against Iran,
09:10the speed was unprecedented. But so was the cost. Reports emerged that one of those strikes hit a
09:18girls' school adjacent to a naval base, killing over a hundred students. Early investigations pointed to
09:25outdated targeting data that had been sitting in the system for years without anyone verifying it.
09:31The weapon hit exactly where it was told to hit. The AI processed the coordinates perfectly. The failure
09:39was upstream and the data nobody had checked. That's the tension at the heart of this. AI doesn't get tired
09:47at three in the morning. It doesn't misread a map or lose focus after 16 hours on shift. But it
09:54also
09:54doesn't question bad data. It doesn't look at a set of coordinates and wonder if there's a school there now.
10:01It processes what it's given and moves to the next target. The friction that existed in the old system,
10:08the phone calls, the sticky notes, the slow human deliberation, that friction also caught errors.
10:16Remove the friction and you remove the safety net along with it. From 743 minutes to strike one target in
10:25a
10:25training exercise, to 15 targets in 60 minutes in a real operation, to 1000 in a single day of combat.
10:33That progression took just 5 years. And the technology is still accelerating. We're only at the
10:40beginning of figuring out what that means. There's more to this story. YouTube's already picked your
10:46next one. It's on screen.
10:48It's on screen.
10:48It's on screen.
10:49It's on screen.
10:49It's on screen.
10:49It's on screen.
Comments