“Future combat will be largely robotic. It will be automated,” former Google Chief Executive Eric Schmidt, who now invests in military-drone companies, said recently on stage at an expo. “It will be controlled by the laws of war.” 

Schmidt’s robot prediction draws little dispute. Less certain is whether last century’s rules can handle warfare’s future.

The war in Iran and AI advances have driven home the dizzying implications of military automation for Washington—and for civilians everywhere. The White House is racing to hammer out policies while the tech world and moral authorities chime in. The cacophony is yielding more questions than answers. 

Ukrainian drone maker Skyfall runs an attack drill in central Ukraine earlier this month. At the last moment the drone operator calls out, ‘Cancel attack.’ Photo: Serhii Korovayny for WSJ

President Trump this month issued an executive order on AI in national security that calls for aggressive use of the technology but requires it to operate “in accordance with applicable laws, government policies, and guidance.” 

The order, which directs the Pentagon to update AI rules adopted only three years ago, sought to bring clarity just as the administration amped up a fight with AI leader Anthropic, whose systems the Pentagon uses. That fight exploded in January over who gets to limit applications of AI in combat and surveillance, and escalated over the weekend. 

“The combination of AI and autonomous weapon systems demands an entirely new approach to risk analysis, risk mitigation, and risk acceptance,” said retired U.S. Air Force Lt. Gen. Jack Shanahan, who led the Pentagon’s first big AI application, Project Maven.

Throughout history, innovations including gunpowder, chemical weapons and airplanes have repeatedly rewritten rules of combat. Only the atomic bomb sparked a civilizational dilemma comparable to AI. And as with nuclear weapons, militaries have entered an automation arms race along a path they can’t foresee. 

U.S. Air Force Lt. Gen. Jack Shanahan, left, in 2020. Office of the Secretary of War Public Affairs

Now, with algorithms controlling not just weapons but entire military networks, humans are ceding wartime judgment to machines on an unprecedented scale. What’s even more worrisome to many: Armed forces are making up the rules as they go along.

Maven’s Legacy

AI entered combat in 2017, when Shanahan’s team used it to nab ISIS bombers attacking U.S. forces in Iraq. Maven’s success leveraging algorithms to scan reconnaissance images proved AI’s potential. Then efforts went quiet as the U.S. pulled back from foreign wars. 

Around 2023, drone makers supplying Ukrainian forces ramped up development of AI to lock onto targets, while commanders began to weave it into targeting systems. Israel, after the Hamas militant attacks that October, tapped AI to sift through mountains of intelligence. The Pentagon, meanwhile, deployed AI-based systems to streamline decision-making. China and Russia have also incorporated AI into military systems.

Today, AI-guided weapons can autonomously home in on objectives a controller picks. That selection generally happens when weapons approach a target, involving a few drones with limited firepower. 

Soon, though, swarms of drones will independently cross great distances by air, water or land to hunt down and strike targets without human intervention. And targets won’t necessarily be on a battlefield.

Killing isn’t AI’s only military assignment. Its role is ballooning across all the less-visceral chores that militaries tackle, particularly in giving priority to intelligence for selecting targets. U.S. commanders say they are selecting targets at more than 10-fold the tempo in Iraq. In the Ukrainian National Guard’s Khartia Corps, automation has tripled the pace of missions, said its top drone engineer.

Serhii Korovayny for WSJ

Serhii Korovayny for WSJ

Drone specialists from Khartia Corps working on recent drone models near Kharkiv, Ukraine, in April. Serhii Korovayny for WSJ

As warfare automation increases, its use is being guided more by battlefield objectives than by codified rules of engagement. At one extreme, Russia is deliberately targeting Ukrainian civilians with both cutting-edge and traditional weapons, seeking to sow terror that no digital regulations will curtail. 

Ukraine, in contrast, lacks resources and wants to maximize each strike’s pain for the Kremlin, meaning collateral damage is often wasteful. For Kyiv, AI can boost efficiency.

At a Ukrainian drone unit called Lasar’s Group, where soldiers hunt high-value Russian targets on AI-enhanced computer images, the new technology is just another tool, said a seasoned pilot who goes by the call sign Sid.

“I understand that I’m setting fire to a vehicle with a crew inside,” said Sid. He isn’t bothered that automatic systems keep targets in his drone’s deadly clutches.

“It’s still a person who presses the button,” he said. “It’s a person who decides whether to activate the system or not.”

That role—dubbed “the person in the loop”—is at the crux of fears about combat automation. Where in the loop is that person? What role do they play? Can they keep pace with computers? 

Technology is advancing so rapidly that even the term itself has morphed into “the person on the loop”—a monitor more than a link in a digital chain.

Concerns over human interactions with AI this year prompted Anthropic to seek explicit Pentagon guarantees that its systems wouldn’t be used for mass domestic surveillance or autonomous weapons, sparking administration backlash. 

Pentagon officials say fears are misplaced because weapons aren’t fully autonomous—and letting technophobia hamstring commanders’ use of AI poses a bigger risk.

“You always have the human who will analyze the situation” and make battlefield judgments on safety and tactics, said Defense Department Chief Digital and AI Officer Cameron Stanley, at the AI+Expo, a recent technology jamboree in Washington organized by a foundation created by Eric Schmidt. 

“The most dangerous course of action right now is to stand still and to remain in a human-driven world,” Stanley said. “The one thing that I am very worried about in war is trying to minimize mistakes.” 

Humanity in War

International Committee of the Red Cross legal adviser Noa Schreuer isn’t sold. Standing in the AI expo’s trade-show hall, amid displays from Amazon.com, Meta and Microsoft, she questioned safety precautions for automated weapons.

“Would an autonomous drone abort a mission on its own, for example if a child enters the target area?” she asked.

A dog-like robot outside the International Red Cross Group's booth at the AI+ Expo in Washington in May. DanIEL Michaels/WSJ

Schreuer was staffing the Red Cross stand—the expo’s buzz-kill. Across its mock battle zone spread a big green sign reading “Humanity in War.” On one faux demolished concrete slab hung a poster-sized page from the Geneva Convention, with rules on protecting civilians. On another hung a fake traffic sign for drone operators reading: “Don’t Outsource Your Authority/Maintain human control and judgment.”

Child deaths in AI-age war aren’t an abstract ethical question. Early in U.S. attacks on Iran this year—as Pentagon officials boasted how new technology was letting them identify and hit targets faster than ever before—Iranian authorities accused the U.S. of striking a school, killing more than 160 people, many of them children. The Pentagon is investigating whether U.S. forces hit the school, which sits near an Iranian military compound, and whether AI was involved.

Automated targeting systems have drawn suspicion, and murkiness around their use means they risk getting blamed no matter what happened. If AI systems offered up the school as a target, investigators must understand what went wrong. If targeters didn’t consult their AI tools, which can instantly scan troves of intelligence, questions will focus on why they didn’t—and whether technology could have helped avoid the civilian deaths.

Israel’s use of AI has drawn similar suspicion, when local media in 2024 alleged that the military was using automated systems to select targets with minimal human oversight and in violation of existing international law, killing large numbers of civilians. The military denied the allegations and issued a detailed defense, saying the systems, code-named Lavender and The Gospel, “are merely tools” for intelligence.

They “do not replace the intelligence analyst,” the Israeli military said. 

But even if AI doesn’t supplant intelligence analysts, can it influence them, or soldiers and commanders? 

That question vexes former Royal Netherlands Air Force Apache attack helicopter pilot Roy Lindelauf, who is now a professor of data science in the department of intelligent systems of Tilburg University. Working from a converted 19th century Dutch locomotive-assembly hall, he and colleagues are trying to mesh this century’s technology with military thinking from the last one. 

“There are so many levels to decision-making,” and how AI figures into them isn’t well understood, said Lindelauf, who also teaches at the Netherlands Defense Academy. 

One concern: People tend to trust what computers tell them, a phenomenon known as automation bias that is being reinforced by lifelike and apparently authoritative digital interfaces. “Even if AI is only a tool, how the human mind works should be taken into account to address biases,” he said. 

In other words, while we think we’re controlling AI, it may actually be controlling us. 

Ukrainian drone maker Skyfall tests their AI-powered drone by aiming for a mannequin target in central Ukraine earlier this month. Serhii Korovayny for WSJ

Skyfall pilots demonstrating their drones’ capabilities. Serhii Korovayny for WSJ

Designing responsible AI was the focus of a report former Dutch Prime Minister Dick Schoof launched at the U.N. General Assembly in September. The Netherlands in 1899 hosted a watershed conference, on Laws and Customs of War on Land, that laid the foundation for modern laws of war.

How AI fits into those aging rules puzzles commanders. Estonian defense adviser Eva Sula works with military leaders across the North Atlantic Treaty Organization, who pepper her with questions nobody can answer. A common one: If I make a mistake because of AI, who is responsible?

“A war crime in the digital space? No country can prosecute it,” said Sula. “They don’t have the laws.”

Human and Machine

For now, many practitioners are improvising. 

Ukraine accepts battlefield risks linked to AI because “the ethical framework is just now being developed,” said Danylo Tsvok, chief executive of the Defense Ministry’s AI warfare center, A1.

Ukraine’s fighters are learning automation’s limits amid combat’s unpredictability, said a serviceman who is part of a team developing AI tools to analyze reconnaissance more quickly. AI can’t yet respond to unconventional situations, he said, and for now “waging war is an art—intuition.”

Stanley, the Pentagon digital and AI officer, said he wants to account for the mistakes that humans and machines can each make alone: “What I am trying to implement is the best human-machine team possible.”

But AI experts foresee surging capabilities and are worried. Pope Leo recently issued an encyclical on AI that built on the work last September of a panel of Nobel laureates, tech specialists and other luminaries the Vatican had assembled. They offered principles and red lines, including a plea that “AI systems must never be allowed to make life-or-death decisions, especially in military applications.” 

Illuminated drones light the sky during the 'Grace For The World' concert in St. Peter’s Square in Vatican City. Stefano Spaziani/ZUMA Press

The conclave was part of a bigger Vatican event that concluded with a free concert in St. Peter’s Square and a light-show of more than 3,000 drones in the night sky.

Panel member Marco Trombetti, chief executive of AI-translation company Translated, said the group’s 18 members agreed that “if these things get used for war, they cannot be stopped.”