Project Maven: Inside the Pentagon's AI Warfare Revolution

From Excel spreadsheets to algorithmic warfare

The U.S. military's transformation into an AI-enabled fighting force has a name: Project Maven. Launched in 2017, the initiative has evolved from rudimentary object detection in drone footage to a platform that reportedly helped identify hundreds of targets daily during Ukraine's defense against Russia.

Katrina Manson's new book chronicles this evolution in unprecedented detail, drawing on hundreds of interviews to reveal a program that has remained largely exempt from Freedom of Information Act requests. According to Manson's reporting, Maven's software has already supported U.S. operations in conflicts with Iran and Venezuela in early 2025, with Palantir's CEO claiming the project "really changed the history of the world."

The journey began with Marine Col. Drew Cukor, who witnessed the limitations of pre-AI military intelligence firsthand. In 2001 Afghanistan, his toolkit consisted of Google Earth, Excel spreadsheets tracking al-Qaeda targets, and PowerPoint diagrams. A decade later, Palantir's data management platform proved more effective, synthesizing information that revealed U.S. forces had raided one location 20 times without realizing it.

The AI breakthrough came in 2017 when Cukor secured $120,000 for a San Francisco startup to train a model on unclassified drone videos. The demonstration—placing white dots on screens to mark specific objects—impressed Deputy Secretary of Defense Robert Work enough to greenlight the project.

Early struggles and the Google controversy

Maven's first real-world deployment in Somalia proved disappointing. Special operators found the AI detections so unreliable—wrong more than half the time—that they disconnected the system entirely. Gradual improvements followed, with algorithms eventually analyzing text, photos, security camera feeds, and balloon footage across more than 60 sites by 2018.

That same year brought Maven's most public crisis. Google employees protested their company's involvement, forcing Google to announce AI principles prohibiting weapons and surveillance work. The company declined to renew its Maven contract, though its algorithms continued supporting operations in Afghanistan through 2019. When Apple acquired the successor company maintaining those algorithms in 2020, it too halted the Maven work.

Palantir emerged as the dominant platform provider, developing what became Maven Smart System—a move that "brought Palantir back to life" according to one team member, as its Pentagon relationship had been winding down.

Ukraine and the automation question

Maven's defining moment arrived with Russia's 2022 invasion of Ukraine. The U.S. Army's 18th Airborne Corps began passing "points of interest"—euphemistically named packages containing target coordinates and timestamps—to Ukrainian forces. On peak days, Maven helped identify 267 potential targets. After Ukraine received GPS-guided HIMARS rockets, striking Maven-identified targets became significantly easier.

Today, Maven operates across all U.S. military branches and multiple combatant commands. NATO acquired its own version of Maven Smart System from Palantir in spring 2025.

Yet the technology raises profound questions about autonomy in warfare. While Maven currently keeps humans in the targeting loop, one officer described the process as "Accept. Accept. Accept"—rapidly concurring with algorithmic conclusions. Gen. Chris Donahue, who expanded Maven's Ukraine support, stated bluntly that "ultimately all this stuff will become automated."

Why it matters

The integration of AI into military targeting represents a fundamental shift in how wars are fought, with implications extending far beyond operational efficiency. As Maven's capabilities improve and spread across allied militaries, the gap between AI-assisted and fully automated targeting narrows. The technology's proponents argue it can reduce civilian casualties and friendly fire incidents by catching patterns humans miss. Critics worry that algorithmic warfare could desensitize combatants, enable higher strike volumes, and—in adversarial hands—facilitate atrocities. With China viewed as the primary competitor in an AI arms race, and phrases like "World War III" appearing throughout military planning discussions, these questions carry existential weight for how democracies will wage future conflicts.

The ethical terrain remains contested. Some Maven advocates point to missed opportunities to prevent civilian deaths, arguing better AI could have averted tragedies like the Kabul drone strike that killed 10 civilians. Others question whether efficiency gains truly reduce harm or simply enable more strikes. Business incentives complicate the picture—Palantir and other contractors stand to profit substantially from military AI contracts.

These details were first reported by Katrina Manson in her book on Project Maven, which represents rare in-depth journalism on a classified military program.