A Parade in Beijing Triggered an Alarm in Washington
At a military parade in Beijing last September, President Xi Jinping stood alongside Vladimir Putin and Kim Jong-un and watched Chinese forces demonstrate several models of drones designed to fly autonomously alongside fighter jets into battle. The display was not ceremonial. Pentagon officials, according to three U.S. defense and intelligence officials, concluded that America's program for unmanned combat drones was lagging China's. Russia, too, was assessed to be ahead in building production facilities capable of manufacturing advanced drones at scale.
The U.S. response was immediate. Anduril, a defense technology start-up based in California, began manufacturing AI-backed, self-flying drones modeled on what China had displayed — and did so three months ahead of schedule, with production launched outside Columbus, Ohio. The urgency was intentional. The gap was assessed as real.
That sequence — Chinese demonstration, American alarm, accelerated domestic production — captures the rhythm of the broader race. Many nations have quietly engaged in a contest of one-upmanship over autonomous weapons: drones that strike targets without human command, self-flying jets that coordinate attacks at speeds no human pilot can match, and AI systems that process intelligence and generate airstrike recommendations in near real time. The United States and China are at the center. But Russia, Ukraine, India, Israel, Iran, and much of Europe are also investing, building, and watching.
"Russia, China and the United States are all building A.I. arms as a deterrent and for 'mutually assured destruction.'"
Palmer Luckey, founder of Anduril — New York Times, April 12, 2026The comparison to the nuclear age is not rhetorical flourish. Analysts and officials have drawn it directly: as the atomic bomb forced rival nations into an uneasy standoff, AI-backed autonomous weapons are constructing a new deterrence architecture. But the nuclear era at least had decades of arms control theory behind it. The AI era is being built without a framework, without agreed definitions of what a violation would mean, and without any clear endpoint.
Drones That Choose Their Own Targets, Jets That Fly Themselves
China and Russia are experimenting with allowing AI to make battlefield decisions independently, according to two U.S. officials. China is developing systems in which dozens of autonomous drones coordinate attacks without human input. Russia's Lancet drone — initially human-piloted — has been progressively upgraded with autonomous targeting features and deployed continuously in Ukraine, where four years of combat have served as the world's most consequential AI weapons proving ground.
Ukraine: The World's AI Weapons Laboratory
Hobbyist racing drones evolved into systems more lethal than artillery. Remote-controlled boats neutralized Russia's Black Sea fleet. Russia's Lancet drone acquired autonomous targeting mid-war. Ukraine has since begun sharing its battlefield data with Palantir and other firms so that AI systems can learn directly from live combat. Europe's major powers drew their own lesson: in February 2026, Germany, France, Italy, Britain, and Poland announced a joint AI-enabled air defense development program.
China's manufacturing scale adds a dimension that no benchmark comparison of AI models captures. Pentagon officials acknowledge that China can produce autonomous weapons at volumes the United States currently cannot match. Its civil-military fusion strategy — which legally obligates commercially developed AI and sensor technologies to remain accessible to the People's Liberation Army — has drawn private firms into military procurement with no American equivalent in scope. At the 2024 Zhuhai Airshow, state manufacturer Norinco revealed an AI-controlled brigade: armored vehicles and drone swarms coordinated by a single autonomous system. A separate exhibitor showed a 16-ton jet-powered drone designed as a flying aircraft carrier capable of deploying dozens of smaller drones mid-flight.
In 2017, Putin declared that whoever leads in AI "will become the ruler of the world." Xi said in 2024 that technology would be the "main battleground" of geopolitical competition. In January 2026, Defense Secretary Pete Hegseth directed all branches of the U.S. military to adopt AI, telling them to "accelerate like hell." The intentions of all three major powers are not in question. The precise state of their capabilities remains, by design, opaque.
'Left Click, Right Click': Claude, Maven, and the Iran Targeting Machine
A week after American and Israeli forces struck Iran in February, a senior Pentagon official offered a rare public window into what computerized warfare now looks like in practice. At a conference livestreamed by Palantir, the official described a system in which a satellite feed showed a warehouse. A mouse click selected a row of trucks parked outside. Within seconds, the AI software suggested a weapon, calculated fuel and ammunition requirements, weighed cost, and generated a complete strike plan. The human operator's contribution to the entire sequence amounted, in the official's own words, to "left click, right click, left click."
That system is Project Maven — the Defense Department's AI targeting initiative, first conceived in 2017 by Lt. Gen. Jack Shanahan to process the overwhelming volumes of drone footage and satellite imagery that the military was collecting but could not analyze fast enough to be operationally useful. Google was initially recruited to help build it, but employee protests over the ethics of AI-assisted targeting forced the company to withdraw. Palantir took over in 2019.
Embedded with a military version of Claude — the AI made by Anthropic — Maven helped generate thousands of targets in the opening weeks of the Iran campaign, a pace that U.S. Central Command attributed in part to "advanced A.I. tools."
New York Times, April 12, 2026Georgetown University senior fellow Emelia Probasco offered a more measured take. After analyzing thousands of People's Liberation Army procurement documents, she found that China was building systems that mirror American ones — including an apparent effort to replicate the Joint Fires Network, the American program linking sensors and weapons globally so that a drone on one side of the world can cue a strike from the other. Her assessment of Maven's capability: "It's not rocket science. I suspect that China already has something like it."
The Pentagon vs. Anthropic
The Trump administration's push for AI weapons has taken on, in the Times's description, an almost evangelical fervor. Last month, the Pentagon labeled Anthropic — the maker of Claude, embedded in Maven — a security risk, in part because the company sought to place limits on how its technology could be used for automated targeting. The conflict between Anthropic's safety constraints and the military's demand for frictionless autonomous strike capability is one of the defining institutional tensions of the current arms race.
China Is Collecting Data on American Weapons in Real Time
Beyond the direct U.S.-China competition, the Iran conflict has produced a second-order intelligence dynamic that analysts describe as equally significant. Chinese private technology companies — operating under civil-military fusion rules that keep their outputs available to the PLA — have been using AI to analyze open-source satellite imagery, flight tracking data, and shipping information to map American deployments and operational patterns throughout the conflict.
Jing'an Technology claimed its platform tracked four U.S. B-2A Spirit stealth bombers during the opening strikes on Tehran, reconstructing flight paths and intercepting communications. Analysts from Kharon, a firm founded by former Treasury counterterrorism officials, concluded that the most dramatic of these claims likely derived from publicly available aviation channels and historical pattern inference rather than any penetration of classified systems. But regardless of which specific claims hold up, the structural fact is significant: China's AI-enabled intelligence apparatus is being trained against live American military operations, collecting performance data on U.S. weapons systems, radar signatures, and operational doctrine. The intended beneficiary of those lessons is PLA planning for a potential future conflict in the Taiwan Strait.
Russia's Helium Card and the Hidden Front
The Iran war's effects on AI competition extend beyond the battlefield itself. Iranian strikes on Qatar's Ras Laffan industrial complex eliminated roughly one-third of global helium supply — a gas without which chip fabrication equipment cannot operate. Russia's Gazprom, running what has become the world's largest helium complex near the Chinese border, stepped in to fill the gap. Memory chip prices roughly doubled in Q1 2026. TSMC's advanced packaging capacity for Nvidia GPUs sold out months ahead of schedule. China's chip-dependent AI programs face measurably less disruption than those of the United States — a supply chain advantage with direct implications for the pace of the arms race itself.
The Man Who Started Project Maven Says It Keeps Him Up at Night
Lt. Gen. Jack Shanahan helped create Project Maven in 2017. He is now a fellow at the Center for a New American Security, and he is worried. "There is a risk of an escalatory spiral where we're in danger of fielding untested, unsafe and unproven systems if we're not careful," he said, "because we each feel like the other side is hiding something from us." His prescription: governments must set clear boundaries before the technology outpaces their ability to govern it. He has not specified what those boundaries should be, because no one has agreed on them yet.
The fear is grounded in research. In exercises dating to 2020, RAND Corporation analysts explored how autonomous systems could accelerate escalation and erode human decision-making. In one scenario, a U.S.-Japan system responded to a North Korean missile launch by autonomously firing an unexpected counterattack. The RAND report's finding was direct: the speed of autonomous systems produced inadvertent escalation. The machines moved faster than human judgment could intervene.
"A.I. is a general-purpose technology like electricity. And we don't talk about an electricity arms race."
Michael Horowitz, former Pentagon official — New York Times, April 12, 2026The ethical questions that once slowed development — whether machines should be permitted to make lethal decisions without human authorization — have been functionally overtaken by competitive pressure. The only formal agreement between China and the United States on AI weaponry is a nonbinding 2024 pledge to keep human control over nuclear launch decisions. Russia has made no equivalent commitment. Every other domain of autonomous weapons development is ungoverned, with no treaty framework, no verification regime, and no agreed line that constitutes a violation.
Anduril's Palmer Luckey argues that the buildup itself may prevent major war, replicating Cold War deterrence logic: if both sides understand what the machines can do, neither will risk finding out. But deterrence assumes rationality, and the systems being built are explicitly designed to act faster than rational deliberation allows. The race that General Shanahan helped start is now running on its own logic. The people who built the starting gun are no longer certain what they are trying to stop.