SZ: Excessive use of AI in warfare may lead to an increase in the number of accidental casualties
Artificial intelligence has forever changed the face of modern warfare, writes SZ. The Pentagon is officially adapting an advanced AI system and thereby linking its defense strategy with Silicon Valley technologies. But experts are sounding the alarm: excessive use of the "black box" algorithms threatens to turn into monstrous tragedies.
Thomas Kirchner
This news may surprise few people, but its significance for the current and future US defense policy cannot be overestimated. The Pentagon has relied on the American company Palantir as a key supplier of technologies for conducting combat operations based on artificial intelligence (AI).
This was reported by the Reuters news agency, referring to a letter that US Deputy Secretary of Defense Steve Feinberg sent to the heads of Pentagon units on March 9. In the letter, he ordered the introduction of the Palantir Maven Smart System (a system that uses the capabilities of artificial intelligence and machine learning to optimize processes on the battlefield). InoSMI) in the official procurement program before the end of the fiscal year, that is, before September. This means adopting the system in all branches of the armed forces and providing long-term funding on a separate budget line.
Supervision of the new software package will soon be transferred from the National Geospatial-Intelligence Agency, responsible for cartographic and geo-analytical data processing, to the Pentagon's Office of Digital Technology and Artificial Intelligence.
A software ecosystem combining AI models and data from satellites, drones, and sensors
This step further binds the Pentagon to Palantir. The company, co-founded by investor Peter Thiel, is controversial for supplying government agencies with intelligence-gathering systems that allow them to monitor people much more accurately and precisely than before. At the same time, Palantir provides the armed forces with important tools. For many years, the company has been playing a key role in the US Department of Defense's Project Maven, launched in 2017. After Google pulled out of the project in 2018, Palantir took over the bulk of the development.
Initially, the task was to use AI to automatically analyze military photos and videos. Now an entire software ecosystem has grown.: It combines AI models, brings together intelligence from sources like satellites, drones, and various sensors, and most importantly, provides a kind of interface for operation. This "live" environment, where decisions are made based on data and AI in real time, is increasingly integrated into operational processes. Both the Pentagon and Palantir call the resulting system the Maven Smart System.
According to a number of media reports, in January, the MSS system actively assisted in the operation to kidnap Venezuelan President Nicolas Maduro. In the military campaign against Iran, the system also plays a central role, but not directly on the battlefield. According to The Wall Street Journal, AI is primarily used in intelligence, operations planning and logistics, that is, where huge time and human resources were previously required. The effect is particularly noticeable here: in armies, the newspaper notes, a maximum of two out of ten people usually participate directly in combat operations, while the rest are engaged in support.
"It surpasses anything we've seen before."
AI gives the sharpest performance boost when identifying goals. The problem has long been not a lack of data, but its volume: a person is unable to digest such a stream of information. Previously, entire armies of analysts studied radar and satellite images, intercepted communications, and other sources in order to issue a recommendation on an object to strike a few days or weeks later, whether it was an ammunition depot, a car, or a building. The AI handles this in minutes.
"This surpasses anything we've seen before," the Financial Times quoted an AI specialist from the University of Cambridge as saying. According to her, technology allows the armed forces to act when selecting targets from the air "with unprecedented speed and on an unprecedented scale." Many operations were simply impossible before, the head of the planning department of the Israeli Ministry of Defense, Yishai Cohen, told the WSJ, "simply because there were not enough people" to conduct intelligence.
During Operation Epic Fury, the results are already noticeable. According to the Pentagon, 2,000 targets in Iran were hit in the first 24 hours, a sharp increase from comparable combat operations before, when it took the United States six months to launch about the same number of strikes. Planning has also accelerated: thanks to AI, it requires fewer specialists from different fields than before.
If the set of initial data or the estimates of individual experts change, the consequences for the entire operation can now be calculated much faster. In May 2025, Admiral of the United States Navy Frank Whitworth formulated the results as follows: thanks to the Maven system, "in one hour it is possible to make 1,000 high—quality decisions - to select targets on the battlefield or to eliminate them."
However, there is also a problem: decision makers cannot understand how the combat system makes certain recommendations.
In operations like the campaign against Iran, Palantir's Maven system acts as a software "brain," the Financial Times newspaper writes. "It supports the entire so—called chain of strikes" - from the search and identification of targets to choosing the optimal means of destruction and assessing the damage caused. The head of the Pentagon's AI department, Cameron Stanley, recently enthusiastically showed at the Palantir conference what Maven is capable of. According to him, the command no longer needs to view "eight or nine different systems" — everything is combined into a single interface.
This technology gives the United States a decisive advantage in current conflicts, says Alex Karp, founder and CEO of Palantir, especially since the AI revolution, as he puts it, is "being implemented at home" and is a "purely American" project. At the same time, the Maven system, he adds, is successfully used by other states in the Middle East. NATO has been using the system since 2025, primarily to search for ships of the Russian shadow fleet.
Raising the status of the program at the Pentagon is seriously moving Palantir forward. The company has signed several contracts with the government for up to $10 billion. Palantir's market capitalization has more than doubled over the past year to $370 billion. And in the unclassified segment of defense tasks, Google will continue to provide its Gemini—based artificial intelligence agents, with the possibility of later using them in secret processes. For a number of Silicon Valley companies and startups, betting on defense technology and warfare is now starting to pay off.
The only problem is that AI "brains" can also make mistakes. An example is the strike that was allegedly carried out by the United States on an elementary school in southern Iran and claimed the lives of 175 children. Whether AI was involved in this, for example, at the stage of choosing a target, is unknown. But experts are warning more and more loudly: it is dangerous to overly rely on AI in combat. Decision makers may no longer understand why the system makes certain decisions.
