On this page
Defense, Geopolitics, and Power
In a converted warehouse in San Francisco's industrial district, Mike LeBlanc stands beside a matte-black humanoid robot, its featureless face embedded...
In a converted warehouse in San Francisco's industrial district, Mike LeBlanc stands beside a matte-black humanoid robot, its featureless face embedded with camera lenses where eyes should be. LeBlanc is a former Marine who served three deployments to the Middle East. He knows what combat looks like, what it costs. The robot beside him, called Phantom, stands five feet nine inches tall and weighs 175 pounds. It can carry 44 pounds of payload. That payload could be supplies, surveillance equipment, or weapons.
"Don't send a Marine where you can send a bullet first," LeBlanc says, quoting the old military aphorism. Then he updates it for the 21st century: "Don't send a Marine where you can send a robot first."
Foundation Robotics, the company LeBlanc co-founded, has done what no other American humanoid manufacturer has been willing to do. While Boston Dynamics, Figure AI, and Tesla have publicly pledged to avoid military applications, Foundation has embraced them. The company has already signed $10 million in defense contracts with the US Air Force, Navy, and Army. Its stated goal: 10,000 humanoid robots by 2026, 50,000 by 2027. Many of them intended for war.
The age of the humanoid soldier has begun. This chapter is not an endorsement of that reality, but an attempt to understand it.
The Inevitability Question#
Whether military humanoids are desirable is a separate question from whether they are inevitable. The honest assessment is that their development is almost certainly unstoppable, driven by forces that transcend any single nation's policy choices.
The strategic logic is straightforward. A highly capable autonomous humanoid would excel as a soldier, both on front lines and in logistics. Unlike human soldiers, it would need no years of training and would not suffer from fatigue or hesitation in combat. When destroyed, no family would receive a folded flag. For military planners focused on capability and casualty avoidance, the appeal is obvious.
The game theory is equally clear. If one nation develops effective military humanoids, others must follow or accept strategic disadvantage. This dynamic has driven every major weapons development in history, from crossbows to aircraft carriers to nuclear weapons. Humanoid robots will be no different.
The technology is also largely dual-use, meaning the same capabilities that make a robot useful in a factory also make it useful on a battlefield. A humanoid that navigates warehouses can also navigate rubble; the same arms that lift boxes can lift ammunition. The components are interchangeable, and only the application differs. This makes restricting military development without crippling civilian applications nearly impossible.
Perhaps most significantly, the development is already underway. Foundation Robotics is the most visible American example, but intelligence assessments suggest programs in China, Russia, and other nations are advancing rapidly. In November 2025, the Chinese People's Liberation Army publicly demonstrated a humanoid robot that mirrored a soldier's movements in real time, showcasing technology that could enable remote combat operations. The demonstration was held before military representatives from thirteen countries, and no one in the room missed the point.
Foundation and the Breaking of the Taboo#
For years, a tacit agreement held among leading humanoid robotics companies: no military applications. Boston Dynamics' terms of service explicitly prohibited weaponization of its robots. Figure AI published a "master plan" in 2022 pledging to never place humanoids in defense applications. Even Elon Musk, when seeking Chinese components for Tesla's Optimus robots, had to assure Beijing the technology would not be used for military purposes.
Foundation shattered this consensus. Founded in 2024, the company was built from the start with defense applications in mind. Its leadership includes LeBlanc, whose three combat deployments informed the company's strategic direction, and Sankaet Pathak, who serves as CEO. The company acquired a small Florida robotics firm called Boardwalk Robotics, renamed its humanoid from "Alex" to "Phantom," and began aggressively pursuing Pentagon contracts.
The Phantom MK-1 is designed explicitly for harsh environments and combat conditions. Unlike competitors that rely on complex sensor arrays, Phantom uses primarily cameras, a deliberate choice intended to simplify data integration and improve reliability in battlefield conditions. Its proprietary cycloid actuators provide strength and quiet operation. Its broad shoulders and utilitarian frame, painted in tactical black, evoke science fiction battle droids more than factory assistants.
Foundation's business model centers on leasing rather than selling its robots, at approximately $100,000 per unit per year. This approach allows the company to maintain control over its technology while generating recurring revenue. For military applications, leasing also addresses security concerns about sensitive technology falling into adversary hands. The manufacturing cost per unit is approximately $150,000 in current small volumes, a figure the company expects to halve as production scales into the thousands.
The applications Foundation envisions range from logistics to direct combat support. In logistics roles, Phantoms could carry supplies, ammunition, and equipment across terrain too dangerous for human soldiers. In reconnaissance, they could scout buildings, caves, and hostile territory before human units advance. In combat support, they could provide cover fire, breach doors with explosives, or serve as what LeBlanc bluntly calls "a bullet sponge instead of a soldier."
The company has already demonstrated its robots to multiple government agencies. Representatives have met with the Department of Homeland Security to discuss potential border security applications. The Department of Defense has shown interest in logistics, planning, and aircraft maintenance roles. LeBlanc has pitched the robots as security guards for American diplomats, as sentries for sensitive installations, even as protective details for officials traveling in hostile regions.
The company emphasizes that Phantom will operate with human-in-the-loop control, meaning human operators will retain authority over lethal decisions while the robot handles movement, navigation, and reconnaissance. This model mirrors current drone warfare, where pilots use autopilot for routine operations but assume direct control for engagement decisions. LeBlanc has stated that co-founder Pathak is "very concerned" about autonomous weapons and has pledged to remotely disable any robot that operates outside approved parameters.
Foundation's production plans are aggressive: 40 robots in 2025, 10,000 in 2026, 50,000 by 2027. If even a fraction of these projections materialize, the US military will have access to humanoid soldiers at scale within years.
The Race with China#
The competition between the United States and China over military robotics has intensified dramatically in recent years, with each side investing heavily in autonomous systems that could reshape the balance of power.
China's approach has been systematic and state-supported. Between 2020 and 2024, Chinese entities filed 5,688 patents related to humanoid robotics, compared to 1,483 from the United States. The "Made in China 2025" initiative, backed by approximately $300 billion in government subsidies, explicitly targets advanced robotics as a strategic industry. Chinese military doctrine, under the concept of "intelligentized warfare," envisions artificial intelligence and autonomous systems as central to future conflict.
The PLA has already demonstrated operational military robots. Armed quadrupeds, sometimes called "wolf robots," have appeared in joint exercises with Cambodia and other partner nations. In one widely circulated video from 2024, a drone dropped an armed robot dog onto a rooftop to simulate infiltration. Chinese state media has reported on reconnaissance robots working alongside infantry, autonomous vehicles conducting combat support operations, and AI systems capable of analyzing thousands of battlefield scenarios in seconds.
The humanoid demonstration at the November 2025 International Army Cadets Week represented a new threshold. The robot, controlled by a soldier wearing a lightweight motion-sensing suit, replicated its operator's movements in real time, from punches to defensive stances. AI systems ensured the robot mirrored its human controller with notable accuracy, responding almost instantaneously to changes in posture and position. The demonstration was held at the PLA Army Engineering University in Nanjing, with cadets from China and thirteen foreign countries observing. Alongside the humanoid, Chinese forces showcased mine-detection robots, voice-controlled bomb disposal units, and AI-enhanced virtual training platforms.
The technology suggests China is developing telepresence systems that could allow operators to control humanoid soldiers from safety while the robots engage in close combat. Military commentators have noted that such systems could reshape combat doctrines. Operators could conduct precision operations in environments too dangerous for human soldiers, while still providing the judgment and split-second adaptability that fully autonomous systems lack.
Chinese state media has also reported on the development of AI-powered war planning systems. Using models similar to DeepSeek, these systems can reportedly assess thousands of battlefield scenarios, covering different terrain configurations, force deployments, and variable sets, in under a minute. The goal is to accelerate the military decision-making cycle so that Chinese commanders can respond faster than their adversaries can adapt.
The Pentagon has taken notice. The US Army's fiscal 2025 budget included $33 million for initial human-machine integration capabilities, with plans to field the first Human-Machine Integrated Formations by 2027. The Army's Project Convergence exercises have tested robotic combat vehicles, autonomous ground systems, and coordination between human and machine units. The Pentagon's Replicator initiative aims to field thousands of autonomous systems rapidly.
Defense analysts assess that China and the United States are roughly equivalent in humanoid research and development capabilities, but China maintains advantages in manufacturing capacity and component supply chains. This disparity could prove significant if the competition moves from prototype development to mass production.
The Transformation of Warfare#
The telepresence model envisions humanoid robots as remote bodies for human operators. An operator in a command center thousands of miles from the battlefield would experience the robot's surroundings through its cameras and microphones, controlling its movements via haptic interfaces. The robot absorbs the physical danger while the human retains judgment and decision-making authority. This approach addresses many ethical concerns about autonomous weapons while still keeping soldiers out of harm's way.
Current teleoperation technology, already demonstrated by companies like 1X Technologies and Sanctuary AI, allows surprisingly immersive control. Operators report that after brief adjustment periods, they experience a genuine sense of presence in the robot's location. The challenge is latency: any delay between operator input and robot response creates disorientation and reduces effectiveness. For combat applications, where milliseconds matter, minimizing latency while maintaining secure communications presents significant technical challenges.
The hybrid army model integrates humanoid robots with human soldiers in combined units. Robots would handle reconnaissance, logistics, and high-risk tasks like building clearance, while humans would provide command, judgment, and decisions about engagement. The robots would serve as force multipliers, allowing smaller human units to accomplish missions that would otherwise require larger forces. Military planners have suggested ratios from one robot per squad to formations where robots outnumber humans.
The model works because the strengths are complementary. Robots can take punishment, operate continuously, and be replaced without grief. Humans bring the kind of judgment that no AI system can yet replicate: reading a room, accepting a surrender, deciding when not to shoot. In urban combat scenarios, a humanoid robot could enter a structure first, using sensors to detect threats and clear rooms, while human soldiers follow to make complex decisions about civilian presence, surrender acceptance, and proportional response.
The US Army's Human-Machine Integrated Formations initiative explicitly pursues this approach. Initial deployments planned for 2027 will pair soldiers with ground robots for reconnaissance and logistics. Subsequent iterations will expand to attack missions, direct fire suppression, and autonomous tactical maneuver. The progression is deliberately incremental, allowing doctrine and training to evolve alongside technology.
The autonomous swarm model, further in the future but actively researched, would deploy large numbers of coordinated robots operating with minimal human supervision. Individual robots would make tactical decisions based on AI systems and sensor inputs, with humans providing strategic direction rather than operational control. This model offers maximum force multiplication but raises the most serious ethical and legal concerns.
The tradeoffs are real. Telepresence maintains human control but requires robust communications infrastructure that adversaries could target. Hybrid armies combine complementary strengths but demand extensive retraining and doctrinal overhaul. Autonomous swarms offer overwhelming force multiplication but challenge fundamental principles of human accountability in warfare.
The most likely near-term development is a gradual progression from telepresence through hybrid operations, with full autonomy remaining restricted by both technical limitations and policy constraints. Foundation's Phantom is designed for this evolutionary path, capable of remote operation today while built for increasing autonomy as technology and regulations permit.
The ongoing conflict in Ukraine offers a preview of how autonomous and semi-autonomous systems transform warfare. Low-cost Ukrainian first-person-view (FPV) drones, costing between $500 and $1,000, have destroyed Russian tanks worth millions of dollars, fundamentally altering battlefield economics. In a July 2025 Pentagon memorandum, Defense Secretary Pete Hegseth stated that "drones are the biggest battlefield innovation in a generation, accounting for most of this year's casualties in Ukraine."
The lessons are being studied intensively. Small, cheap, numerous systems can overwhelm expensive defenses, and autonomous navigation lets them operate even when communications are jammed. A single operator controlling multiple drones simultaneously achieves force multiplication impossible with human soldiers alone. These principles apply directly to humanoid robots, suggesting that future ground warfare may be characterized by large numbers of expendable robotic units rather than smaller numbers of highly protected human soldiers.
The Ethical Battlefield#
The development of humanoid military robots forces confrontation with questions that military ethics has never fully resolved, even regarding human soldiers. Autonomous systems make these questions more urgent and more difficult.
International humanitarian law requires distinction between combatants and civilians, proportionality between military necessity and civilian harm, and precaution to minimize unintended damage. These principles assume human judgment at critical decision points. A soldier can recognize a civilian who has wandered into a combat zone and decide to hold fire. A commander can weigh collateral damage against military necessity, or choose to accept a surrender that an algorithm might not recognize as such.
Whether autonomous systems can replicate these judgments is contested. Proponents argue that robots with sophisticated sensors and AI could potentially make more accurate distinctions than stressed, fatigued human soldiers operating under fire. Opponents counter that moral judgment requires capacities that machines lack: understanding of context, appreciation of human dignity, and accountability for consequences.
The accountability question is perhaps most troubling. When a human soldier commits a war crime, they can be identified, tried, and punished. When an autonomous system kills civilians, who bears responsibility? The programmer who wrote the targeting algorithm? The commander who authorized deployment? The diffusion of responsibility across complex technical and institutional systems may leave no one effectively accountable.
Human Rights Watch and other organizations have argued that autonomous weapons would violate the principle of human dignity by reducing people to data points in a targeting algorithm. As Bonnie Docherty, a lecturer at Harvard Law School, has stated: "Autonomous weapon systems cannot have that human capacity to understand the value of a life or the significance of its loss, because they are not living beings."
The counterargument, advanced by some military ethicists, is that autonomous systems might actually reduce overall harm. If robots can accomplish military objectives with less collateral damage than human soldiers, if they can eliminate human error, fatigue, and revenge killing, their deployment might be more ethical than alternatives. This utilitarian calculus has appeal but remains deeply contested.
The International Response#
The international community has debated regulation of autonomous weapons systems since 2014, with limited concrete results. The United Nations Convention on Certain Conventional Weapons has hosted annual discussions, but its consensus-based decision-making has allowed a handful of major military powers to block binding agreements.
In December 2024, the UN General Assembly adopted a resolution on lethal autonomous weapons systems by a vote of 166 in favor and 3 opposed, with 15 abstentions. Only Belarus, North Korea, and Russia voted against. The resolution expressed concern about autonomous weapons' impact on international peace and security, including risks of arms races, conflict escalation, and proliferation to non-state actors. A subsequent resolution in November 2025 passed with 156 votes in favor and only 5 opposed.
The resolution called for negotiations toward a legally binding instrument and endorsed what advocates call a "two-tiered approach": outright prohibition of autonomous weapons that operate without meaningful human control or that target people, combined with regulations ensuring human control over all other autonomous weapons systems.
UN Secretary-General Antonio Guterres has been unequivocal in his position. "Machines that have the power and discretion to take human lives without human control are politically unacceptable, morally repugnant, and should be banned by international law," he stated in May 2025. He has called for a binding international agreement by 2026.
Despite this momentum, major military powers have resisted binding commitments. The United States, Russia, India, and Israel have consistently opposed proposals that might constrain their weapons development. The US Department of Defense directive on autonomous weapons, issued in 2012 and updated since, requires human oversight but does not prohibit development or deployment of increasingly autonomous systems.
The pattern mirrors historical arms control efforts. International agreements tend to emerge only after weapons have proliferated enough to make their dangers undeniable, if they emerge at all. The Campaign to Stop Killer Robots, a coalition of over 270 civil society organizations, continues advocating for preemptive regulation, but the window for prevention may already be closing.
Geopolitical Implications#
Military humanoids will not affect all nations equally. Their development is likely to concentrate power among wealthy, technologically advanced nations while creating new vulnerabilities for others.
The capital costs of developing and manufacturing advanced humanoid robots are substantial. Foundation estimates its Phantom costs approximately $150,000 to manufacture in small volumes, though this could decrease significantly at scale. Research and development costs are far higher. Only nations with robust industrial bases, advanced AI capabilities, and substantial defense budgets can compete in the initial development phase.
This concentration of capability could destabilize existing power balances. Smaller nations that have relied on human military forces to maintain deterrence may find themselves at systematic disadvantage against adversaries with large robot armies. The traditional assumption that military power correlates roughly with population may erode if humanoid soldiers can be manufactured rather than trained.
Regional powers face difficult choices. Nations like Japan, South Korea, and European states may need to decide whether to develop independent military humanoid programs, purchase or lease systems from the United States, or accept strategic vulnerability. Alliance relationships may shift as the military significance of human troop contributions declines relative to technological capabilities.
Non-state actors present particular concerns. The same manufacturing capabilities that could produce tens of thousands of military humanoids could, in time, be accessed by terrorist organizations, cartels, or other groups outside state control. The proliferation risk parallels earlier concerns about nuclear weapons, but with potentially lower barriers to acquisition as costs decline.
Cybersecurity vulnerabilities add another dimension of risk. Military humanoids would be deeply integrated with communications networks, AI systems, and command structures. A successful cyberattack could disable robots at critical moments, feed false sensor data to mislead operators, or potentially turn robots against their own forces. Michael Vermeer, a senior physical scientist at the RAND Corporation, has warned that "cybersecurity and electronic warfare defenses would need to be carefully managed to ensure that the humanoid robots could not be deactivated, or worse, used against U.S. forces."
The spectrum management challenge is equally significant. Current military communications infrastructure lacks sufficient bandwidth to support large numbers of teleoperated robots operating simultaneously on the battlefield. As one Army official noted, there is not enough spectrum allocated to military operations for reasonable density of robotic ground combat vehicles using current teleoperation methods. This constraint drives development toward greater autonomy, since robots that can operate independently require less communications bandwidth than those requiring constant human control.
Perhaps most significantly, military humanoids could lower the political barriers to conflict. When citizens' children are not at risk, when casualty reports do not generate domestic opposition, political leaders may find it easier to authorize military action. The restraint that democracies have historically shown when their populations bear the costs of war could erode if those costs become primarily economic rather than human.
Some advocates argue that robot warfare could actually reduce suffering by removing humans from combat zones on both sides. If opposing armies were primarily robotic, conflicts might be decided with minimal human casualties. This optimistic scenario assumes, however, that military humanoids would fight only each other, that civilian populations would be spared, and that conflicts would remain contained. History offers little support for such assumptions.
The Defense Industrial Transformation#
The emergence of military humanoids is restructuring defense industries in ways that advantage newer entrants over traditional contractors.
Traditional defense firms like Lockheed Martin, Raytheon, and General Dynamics built their dominance on decades of experience with complex weapons systems, established relationships with procurement officials, and mastery of government contracting requirements. Their products, aircraft carriers, fighter jets, missile systems, require enormous capital investment and decades-long development cycles.
The humanoid robotics industry operates differently. Companies like Foundation and Anduril Industries, while smaller than traditional defense contractors, move faster and leverage technologies developed in commercial markets. Anduril, valued at $30.5 billion after a $2.5 billion funding round in June 2025, has built its approach around software-first development, using its Lattice platform to integrate diverse systems and enable rapid updates. The company is constructing Arsenal-1, a five-million-square-foot manufacturing facility in Ohio, with production slated to begin in July 2026 and capacity to produce tens of thousands of autonomous systems annually.
Foundation's business model, leasing robots rather than selling them, represents another innovation foreign to traditional defense contracting. The approach allows continuous updates, maintains control over deployed systems, and generates recurring revenue streams that can fund ongoing development without constant contract renegotiation.
This transformation has policy implications. The traditional defense industrial base, concentrated among a handful of established contractors, offered predictability and accountability. The emerging landscape is more diffuse, with startups competing alongside legacy firms and dual-use technologies blurring lines between civilian and military applications. Oversight becomes more difficult when the same robots that staff Amazon warehouses could, with software modifications, patrol forward operating bases.
The manufacturing implications are equally significant. Both the United States and China are investing in domestic production capacity for military robots, recognizing that supply chain dependencies create strategic vulnerabilities. Anduril's Arsenal-1 facility explicitly aims to produce systems faster than adversary manufacturers. China's robotics industry, already dominant in commercial markets, provides a foundation for military production that the United States struggles to match.
What Comes Next#
Military humanoids are coming. The remaining questions are practical: how fast, at what scale, and under what constraints. Can Foundation actually ramp from 40 robots to 10,000? Will the US-China competition accelerate into an arms race, or stabilize at a level both sides can sustain? And will international negotiations produce meaningful constraints before proliferation makes the question moot?
The technical obstacles are real. Autonomous systems operating in contested environments face problems that laboratory demonstrations have not fully solved. Communications links can be jammed, sensors deceived, and AI systems can fail catastrophically when confronted with scenarios outside their training data. The gap between a controlled demonstration and reliable operational capability remains substantial. But policy choices matter more than technological determinism might suggest, and competitive pressure has a way of crowding out careful deliberation.
This chapter has tried to avoid both naive optimism and reflexive alarm. Military humanoids may reduce human casualties in conflicts that would occur regardless. They may also lower barriers to conflict, accelerate arms races, and create accountability gaps that undermine the laws of war. The technology itself is neutral; its consequences will depend on how humans choose to develop and deploy it.
Mike LeBlanc, standing beside his company's robot soldier, believes he is reducing the risk to young Marines who might otherwise be sent into danger. His conviction is genuine, rooted in his own experience of combat and its costs. But conviction alone does not resolve the ethical questions his technology raises. Those questions belong to all of us, because the decisions being made now will shape the nature of warfare for generations. The task is to ensure that when these machines arrive in force, they remain tools in human hands, subject to the restraint and accountability that civilization demands. That work is not primarily technological. It is political and moral, and it has already begun.