The Economist's May 14, 2026 review of Katrina Manson's Project Maven (W.W. Norton, 416 pages, $31.99 / £23) describes the U.S. military's AI-targeting program, which one NATO official quoted in the book calls "the Microsoft Windows of warfighting." Manson, a national-security journalist at Bloomberg, documents how the Maven Smart System fuses intelligence and coordinates strikes at scale, how its corporate base shifted from Google's 2018 walkout to Palantir as anchor contractor, and how accuracy and degrees of autonomy have advanced while human oversight remains the central operational constraint (Source: economist.com).
URL: economist.com Date: 2026-05-14 Book reviewed: Project Maven by Katrina Manson (W.W. Norton, 416 pages, $31.99 / £23). Manson is a national-security journalist at Bloomberg. Class: foundational (review of a foundational primary source — Manson's book — documenting the AI-targeting program of the U.S. military).
What Maven does
According to the review, Maven fuses intelligence types — photographs, text, radio, and electromagnetic pulses — into target identification, then coordinates response options such as which aircraft and which munitions are closest to a given target. The book states that "with a single click by a human, Maven can turn data into ash." A National Geospatial-Intelligence Agency official cited by Manson says the system identifies and hits roughly 5,000 targets per day, which the same official describes as a fivefold speedup over the pre-LLM era; on one day in 2022 in Ukraine it found more than 260 potential targets. Manson reports that the algorithms now run inside weapons rather than only on distant servers, naming Goalkeeper, a loitering munition or suicide drone, and Whiplash, an explosives-laden jet ski whose early versions were tested in Ukraine via the CIA.
Operational history
The review traces Maven from early counter-terror trials through industrial-scale use in Ukraine. In 2017, trials in Somalia produced errors in which algorithms labeled clouds as flying school buses; in 2018, software in Afghanistan identified trees as people and rocks as buildings. Google walked out of Maven in 2018 after employee protests over collaborating on lethal AI. The program was subsequently used in the 2019 operation that killed Abu Bakr al-Baghdadi of ISIS in Syria and in the 2020 drone strike that killed Qassem Suleimani of Iran, and in 2021 it tracked crowd density at Kabul airfield during the U.S. withdrawal.
Russia's 2022 invasion of Ukraine marked the system's shift to industrial-scale targeting, with Maven feeding "points of interest" against a cloud-computing bill of roughly $1 million per month. Manson reports that between late 2022 and 2023 more than 1,500 algorithms were whittled down to about two dozen for Ukraine, and that accuracy proved context-dependent: a model achieving 70% accuracy in Afghanistan dropped to 30% in the Philippines, which the book attributes to jungle versus dusty terrain, and the system struggled with Ukrainian snow and with Russian tanks missing their turrets. By 2026, the review reports, Maven still produced 10 false detections per square kilometer assessed in Ukraine.
On data acquisition, Manson recounts that Navy P-8 sub-hunters wiped their hard drives after each mission until Maven forced data preservation, and that the submarine-hunting training set ultimately drew from "boat cameras, port cameras, infrared systems, destroyers, combat ships, buoys, dhows." An unnamed defense official is quoted: "We're basically watching the PLA all the time… to get AI training data."
People
Drew Cukor, a Marine officer, founded Maven and built what the review calls a "team of mavericks"; he was reprimanded by the Marine Corps for overseeing a toxic culture. Manson quotes him: "Let's be able to look ourselves in the mirror and make sure we are careful. We have all this tech; are we the best custodians of it?" General Chris Donahue, commander of American land forces in Europe, pioneered Maven's use in Ukraine and is quoted saying, "Ultimately, all this stuff will become automated." Against Donahue's framing of an automated trajectory, the review asks "even if people are overseeing the process, can they keep up, particularly if something goes wrong?"
Silicon Valley defense pivot
The review describes a shift in Maven's corporate base. After Google walked out in 2018, Palantir became the most important firm in Maven's development. Manson and the reviewer frame this as part of a broader pattern in which Palantir Technologies, Anduril Industries, and SpaceX displace traditional defense primes, a dynamic covered under Autonomous Weapons. The book characterizes the 2018 protests as a Silicon Valley refusal to militarize AI that, eight years on, slowed Google's own defense business rather than Maven itself, as competitors absorbed the work. The review treats the "Maven Smart System" label as understating the program, arguing that it functions as a kill-chain coordinator rather than an intelligence tool and was built for targeting from the outset.
Key claims
| Claim | Confidence | Notes |
|---|---|---|
| Maven enables 5× faster targeting | high | Manson sourced from NGA official; consistent with public Maven coverage. |
| Maven kill chain used vs. Baghdadi, Suleimani | high | Public record, confirmed in Manson's book. |
| 10 false detections per km² persisted in Ukraine | high | Manson book; specific operational reliability number. |
| LLMs (now part of Maven) "speeded up targeting fivefold" | high | NGA-official quote. |
| Models trained on Afghanistan dropped from 70% → 30% accuracy in Philippines | high | Manson book; empirical case for context-specific accuracy. |
| Goalkeeper and Whiplash carry on-board algorithms | medium | Manson reporting; not all weapon-system details verified independently. |
| General Donahue: "all this stuff will become automated" | high | Direct quote. |
Reception and tensions
The review presents Maven as a case in which operationally consequential AI is already in use rather than a forecast for the 2030s, and treats the 30% Philippines accuracy figure and the 10-false-detections-per-square-kilometer Ukraine figure as concrete reference points for how context-dependent such systems remain. The Economist's portrayal of Cukor as an "intriguing character" sits alongside the toxic-culture reprimand the book documents, a contrast the review acknowledges. The piece notes that Manson's reporting is specific enough to anchor claims about autonomous-weapons capability against Maven's measured performance rather than against speculation, including production-scale use with false-detection rates that operators tolerate.
Relationships
- supports: Autonomous Weapons, AI and National Security, DOD — Department of Defense (AI Deployer), Chief Digital and AI Office (CDAO) (Chief Digital and AI Office adjacency)
- depends-on: Palantir Technologies (current anchor contractor), Anduril Industries (parallel defense-AI development)
- contradicts: Silicon Valley narrative that defense AI is "too unreliable for production use" — Manson documents production-scale use with measurable false-detection rates that operators tolerate
- related: Katrina Manson (Bloomberg national-security journalist; foundational author — entity created), Drew Cukor (Maven founder — entity created), Intelligence Replaces Hierarchy (kill-chain compression mirrors corporate-hierarchy flattening), Post-Deployment AI System Monitoring (10/km² false-positive rate as ongoing concern)
- regulated-by: DOD — Department of Defense (AI Deployer) (Directive 3000.09 on autonomy in weapons), Hague law of armed conflict adjacency
Wiki Folding
Updates Autonomous Weapons (Maven operational anchor + 5K targets/day + 10/km² false-positive rate), AI and National Security, Defense / Military — AI Deployment (Project Maven canonical reference), Palantir Technologies (Maven anchor role), creates Katrina Manson and Drew Cukor entity pages, and links to Anthropic Claude Gov + Pentagon Dispute (2025–2026) for the current frontier-lab–Pentagon dynamic.