Project Maven has become one of the clearest examples of how artificial intelligence is changing military decision-making. What began as an attempt to help analyze imagery has grown into a system described as capable of pulling together many streams of battlefield data and moving the targeting process much faster.
The debate around military AI often focuses on chatbots and large language models. But the source article argues that the deeper shift started earlier, with systems built to find, organize, and act on intelligence at speed.
From drone footage to battlefield workflow
Project Maven started in 2017 as an experiment in using computer vision on drone footage. Journalist Katrina Manson examines that history in her book Project Maven: A Marine Colonel, His Team, and the Dawn of AI Warfare, which follows the system’s development and the people behind it.
The project was pushed forward by Drew Cukor, a Marine intelligence officer. Manson describes him as frustrated by the intelligence tools available to US military operators in Afghanistan. Information was often handled through Excel and PowerPoint, and knowledge could be lost when troops rotated out.
Cukor wanted a more useful analytic system for frontline operators. One part of that vision was what he called “white dots”: points on a map connected to intelligence information such as a coordinate, what is located there, the elevation, and what is known about it.
The initial idea was narrower: use AI to help review drone video footage. Manson says operators were sometimes able to analyze as little as 4 percent of the collection, so AI was seen as a way to reduce the burden on human analysts. But the ambition was never limited to sorting images. The system developed into something closer to a targeting workflow.
Why Google became part of the controversy
The public first encountered Project Maven through employee protests at Google in 2018. Google was the military’s initial contractor, but the protests pushed the company to back out.
At the time, a Google spokesperson said the use of AI to flag images for review on drone feeds was meant to save lives and was for non-offensive uses only. Manson’s reporting presents a more complicated picture. She says many military operators were motivated by saving US lives and reducing civilian harm, but that AI target selection was intended for targeting in a practical sense.
One person quoted in the book put that point directly: “yeah, of course, it’s not like we’re doing it for kicks. The goal of the intel is to take out high-value targets.”
After the Google deal fell apart, Palantir stepped in. Microsoft and AWS took larger roles in algorithms and compute, while Cukor asked Palantir for help building the interface that would make the system usable. The source article says Maven eventually drew on technologies developed by Microsoft, Amazon, Anthropic, and others.
What the Maven Smart System does
The Maven Smart System is described as combining computer vision with a workflow management system. It can find targets, pair them with weapons, and allow users to move through other steps in the targeting cycle quickly.
The system also synthesizes satellite imagery, radar, social media, and dozens of other data sources. It is now used across the US armed forces and has recently been purchased by NATO.
The speed is central to the concern. A process that once took hours can now be completed in seconds. An official tells Manson that the technology helped the US move from hitting under a hundred targets a day to a thousand, and with the addition of LLMs, up to five thousand targets a day.
That change alters more than efficiency. It changes the rhythm of decisions. When more targets can be processed in less time, old errors in databases, imagery, or assumptions can become more consequential because the system can act on them faster.
The Iran strike and the cost of acceleration
The source article opens with the first 24 hours of the assault on Iran, when the US military struck more than 1,000 targets. That was nearly double the scale of the “shock and awe” attack on Iraq over two decades ago. AI systems, especially the Maven Smart System, are described as key to that acceleration.
One of the thousand targets struck on the first day of the Iran war was a girls’ school. More than 150 people were killed, mostly children. The school had previously been part of an Iranian naval base, but it was listed online as a school, and playgrounds were visible on satellite imagery.
Much of the attention after the strike focused on possible hallucinations by Claude. The technology historian Kevin Baker argued in The Guardian that the more important issue was Maven and the acceleration it enabled. He wrote: “A chatbot did not kill those children,” and added, “People failed to update a database, and other people built a system fast enough to make that failure lethal.”
The point is not that speed alone explains every failure. It is that speed changes the consequences of failure. When a targeting system compresses a process from hours into seconds, the quality of every input, update, review, and human decision matters more.
Where military AI goes next
Manson’s reporting also looks beyond Maven. The source article says she uncovered military programs to develop fully autonomous weapons, including an explosive-laden drone Jet Ski, capable of targeting and destroying targets on their own.
That suggests the debate over military AI is moving from assistance to autonomy. Maven already shows how AI can organize information, highlight possible targets, and accelerate a kill chain. The next question is how far systems will be allowed to go without human control at each step.
Project Maven’s importance is that it connects Silicon Valley tools, military workflows, and battlefield consequences. It began with computer vision and drone footage, but its larger impact is in how quickly intelligence can be turned into action.
For readers trying to understand military AI, Maven is a more revealing case than any single chatbot. It shows that the most important AI systems may not look conversational at all. They may look like maps, databases, imagery tools, and workflows that make war move faster.