The Wire · live □ GRID-REF 37°47′N 122°25′W
DISPATCH 04/26 · 20 Nov 2026
BattlePolicy
Startup to front line. Strategy to consequence.
News · USA

Airwars: At Least 79 Civilians Reported Killed in U.S. Somalia Strikes in Maven's First Two Years, Up From 25

The watchdog says its before-and-after count cannot isolate the AI, but it finds no evidence the targeting software reduced harm, in the month a Palantir paper pitching civilian-harm safeguards reached U.S. officials.

Airwars: At Least 79 Civilians Reported Killed in U.S. Somalia Strikes in Maven's First Two Years, Up From 25
FIG.01 · USA Illustration. Generated key image, not a photo of the event.

The watchdog says its before-and-after count cannot isolate the AI, but it finds no evidence the targeting software reduced harm, in the month a Palantir paper pitching civilian-harm safeguards reached U.S. officials.

Reported civilian deaths in alleged U.S. strikes in Somalia more than tripled in the two years after Project Maven's computer-vision models arrived there in December 2017, Airwars found in an analysis published Nov. 18. The watchdog documented a minimum of 25 civilians killed in the two years before and a minimum of 79 after.

Strikes attributed to U.S. forces rose from 64 to 153. The share of alleged strikes with reported civilian deaths or injuries went from roughly one in seven to more than one in five.

Frequently Asked Questions

What did Airwars find?

Airwars documented a minimum of 25 civilians killed in 64 strikes attributed to U.S. forces in Somalia in the two years before Project Maven arrived in December 2017, and a minimum of 79 killed in the two years after, when it counted 153 strikes.

Does the analysis show that AI caused the deaths?

No. Airwars says other factors contributed, including looser strike rules after part of Somalia was designated an area of active hostilities in 2017. Its stated conclusion is that there is no evidence the introduction of AI reduced harm or made targeting more accurate.

What is Project Maven?

Project Maven is the Pentagon's algorithmic warfare program. It trained computer-vision models on drone and other sensor imagery to detect objects and targets. Palantir's Maven Smart System is the interface that fuses the feeds, according to Airwars.

How accurate were the early Maven models?

Katrina Manson's book "Project Maven" says first-generation models deployed in new terrain showed detection accuracy below 30%, and at times under 10%. Airwars cites the figure without tying it to a specific country.

How did Palantir respond?

A Palantir spokesperson told Airwars that Maven's technology is "critical to both deterrence, the avoidance of wars, and the protection of civilians in the case of war."

What is Palantir's civilian harm white paper?

TIME reported that a Palantir document dated November 2026 pitches civilian no-strike data layers, an action log and a legal-review workflow. TIME found it does not appear to include a system for logging and investigating civilian-harm allegations after a strike.

AI-generated summary, reviewed by an editor. More on our AI guidelines.

Read next — from the Lexicon

Below this point

6 more sections.

The comparison does not establish that Maven caused the increase. Airwars lists other drivers. Shortly before fielding, President Donald Trump designated part of Somalia an "area of active hostilities," loosening the near-certainty standard for strikes. How many U.S. strikes Maven's algorithms aided remains classified. The rise in strikes is "consistent with the goals of Project Maven," which include finding more targets, the group added. Its conclusion is narrow: "there is no evidence, as advocates claim, that the introduction of AI reduced harm or made targeting more accurate."

Maven reached Baledogle airfield in December 2017, according to Katrina Manson's book "Project Maven." Manson separately reports that first-generation models showed detection accuracy below 30% in new terrain, at times under 10%. Palantir, which stepped in after Google left in 2018, builds Maven Smart System, the interface that fuses the feeds.

A Palantir spokesperson told Airwars that Maven's "superior technology is critical to both deterrence, the avoidance of wars, and the protection of civilians in the case of war."

The deputy defense secretary announced earlier this year that Maven Smart System would become a program of record, Airwars noted. A Palantir document dated November 2026, titled "Palantir Civilian Harm Mitigation Technical Whitepaper," has circulated among U.S. officials this month, TIME reported. It pitches civilian "no strike" data layers and an action log inside Gaia, the geospatial interface within Maven Smart System, plus a legal-review workflow.

The features do not appear to include a system for logging and investigating civilian-harm allegations after a strike, the magazine wrote. Airwars executive director Emily Tripp called that "a glaring gap."

Nations at the U.N. Convention on Certain Conventional Weapons decide this week whether to mandate negotiations on an autonomous-weapons treaty. Target-identification models such as Maven would sit outside it, per Airwars: the U.N. definition covers only a single system that both selects and strikes a target.

San Francisco, California, USA

Marcus Schuler edits BattlePolicy, a daily defense-technology brief connecting the companies and capabilities behind modern war to the contest among Europe, the US, Russia, and China.

FIELD DISPATCH · WEEKLY

BattlePolicy Weekly — free.

Defense tech, startups, and security — weekly.

Related