Foundation Future Industries, a San Francisco startup founded in April 2024, has secured $24 million in research contracts with the US Army, Navy, and Air Force to test humanoid robots designed to breach enemy positions. The company’s Phantom MK-1 is a 5-foot-9, 176-pound humanoid with 19 upper-body degrees of freedom, five-fingered hands, a camera-first vision system, and an LLM-driven autonomy stack that blends independent operation with supervised teleoperation. Two units were sent to Ukraine in February for frontline testing in logistics and reconnaissance, described as the first deployment of humanoid robots to any theatre of combat. The company is seeking $500 million in new funding at a valuation exceeding $3 billion. Its chief strategy adviser is Eric Trump, the son of the sitting president, a detail that prompted Senator Elizabeth Warren to call the Pentagon contracts “corruption in plain sight.” The company’s CEO previously ran a fintech startup that went bankrupt with tens of millions in consumer deposits unaccounted for.

The machine

The Phantom MK-1 walks at 1.7 metres per second, carries a 44-pound payload, runs on eight cameras with no bulky LiDAR, and uses proprietary cycloidal actuators delivering up to 160 newton-metres of torque. Its AI stack translates high-level task instructions into motion through an LLM pipeline, with operators retaining final authority over lethal decisions. The unit cost is approximately $150,000, with a lease model available at $100,000 per year. The MK-2, expected this month, consolidates electronics to reduce short-circuit risk, adds waterproofing and larger battery packs, increases payload capacity to 175 pounds, and uses cast-moulded bodywork to speed manufacturing and reduce costs. Foundation’s production targets are 40 units in 2025, 10,000 in 2026, and 50,000 by the end of 2027, with a steady-state target of 30,000 per year. Those numbers would require a manufacturing scale-up of 250 times in two years on a total funding base of roughly $21 million.

The company was founded by Sankaet Pathak, previously the CEO of Synapse, a banking-as-a-service platform that filed for bankruptcy in 2024; Arjun Sethi, CEO of Tribe Capital, which led Foundation’s $11 million pre-seed round; and Mike LeBlanc, a 14-year Marine Corps veteran and co-founder of Cobalt Robotics. LeBlanc provides the military credibility and has said the company believes there is “a moral imperative to put these robots into war instead of soldiers.” In June 2024, CNBC reported that Foundation had been fundraising with exaggerated claims about ties to General Motors, including assertions that GM had committed to invest and placed a $300 million purchase order. GM flatly denied all of it. LeBlanc confirmed the denial and said he was “embarrassed” the marketing materials existed. For a company asking the Pentagon to trust its robots in combat, the credibility gap matters.

The contracts

The $24 million in Pentagon contracts includes an SBIR Phase 3 designation, which qualifies Foundation as an approved military vendor, and specific research agreements for testing humanoid robots in breaching scenarios. Some contracts were inherited through the acquisition of a company called Boardwalk, including a US Air Force SBIR award valued at approximately $1.8 million. Eric Trump appeared on Fox Business to tout the contracts. Warren’s response was immediate: “Is the Pentagon a cash machine for Trump’s kids?” The political dimension is unavoidable. A sitting president’s son serving as chief strategy adviser to a company receiving Defence Department contracts raises governance questions regardless of the company’s technical merits. The contracts are real, but they are small. Shield AI recently raised $2 billion to scale its autonomous combat pilot, an AI system called Hivemind that flies aircraft autonomously and has been tested in combat conditions. Anduril secured a landmark $20 billion, ten-year US Army contract in March for its AI-enabled Lattice platform. Foundation’s $24 million is a research agreement, not a production order. The gap between a research contract and a deployed weapons system is measured in billions of dollars and years of testing.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

The Ukraine deployment adds a different kind of credibility. Two Phantom MK-1 units sent for logistics runs and reconnaissance sweeps in February represent real-world testing in a live conflict zone, and Foundation is using battlefield feedback to refine the MK-2 design. But “tested in Ukraine” is not “deployed in combat.” No humanoid robot has fired a weapon in a conflict. The units performed support tasks. The distinction matters because the company’s marketing, its fundraising narrative, and its Pentagon contracts all converge on the idea of a humanoid soldier, and the technology is not there yet. NATO-backed ARX Robotics secured 31 million euros for its autonomous battlefield robots, ground vehicles that perform logistics and reconnaissance without the complexity of bipedal locomotion. ARX Robotics is already scaling production of autonomous land drones to 1,800 units a year at a new UK plant, a manufacturing reality that Foundation’s targets have not yet approached.

The debate

The Campaign to Stop Killer Robots, a coalition of more than 250 NGOs, has been advocating for a new international legal instrument ensuring human control in the use of force since 2013. Approximately 90 states have called for such an instrument. A minority of militarised states, including the United States and Russia, have blocked its adoption. In November 2025, the UN General Assembly First Committee adopted a resolution with 156 states in favour and 5 against calling for negotiations on autonomous weapons. The Group of Governmental Experts on lethal autonomous weapons systems has sessions scheduled for 2026 and is expected to submit a final report to the Convention on Conventional Weapons in November. This is the last year of the GGE’s mandate, making 2026 a make-or-break year for international regulation of autonomous weapons.

Foundation’s stated policy is that human operators retain final authority over lethal decisions, a “human-in-the-loop” commitment that the Pentagon’s own Directive 3000.09 on autonomy in weapon systems requires for autonomous and semi-autonomous platforms. But the company’s LLM-driven autonomy stack and its stated ambition to “reduce teleoperation needs over time” are in tension with that commitment. An LLM-driven task-to-motion pipeline that learns to operate more independently with each iteration is, by design, moving toward the autonomous capability that the international community is trying to regulate. The AI warfare push that made Helsing one of Europe’s most valuable tech firms, valued at 12 billion euros for military AI software that coordinates drone swarms, shows the scale of capital flowing into autonomous military systems. The ethical guardrails are voluntary. The funding incentives point in one direction.

The race

China demonstrated a motion-controlled humanoid robot for military tasks at an international military cadets event in Nanjing. WuBa Intelligent Tech secured approximately $69 million for its RoboWolf quadrupeds, backed by NORINCO, the state-owned defence conglomerate. The Pentagon added Unitree, a consumer robot-dog maker, to its Chinese Military Companies list in February 2026. War on the Rocks reported on a hidden system turning Chinese technology companies into military suppliers. Viral videos purporting to show a Chinese humanoid robot army were debunked by France 24 as AI-generated fakes, but the fakes themselves reflect the narrative arms race: the perception that a country is building robot soldiers may matter as much as the reality in shaping defence budgets and procurement decisions.

Russia has established an Unmanned Systems Forces as a new military branch, is deploying the Kurier autonomous mortar system that loads and fires without human input, and is rapidly expanding its ground drone fleet in Ukraine. Neither country has deployed humanoid robots in combat. The military robotics that are actually in use, on both sides of the Ukraine war and in US border patrol and base security operations, are wheeled, tracked, or quadruped. They succeed because they are simple, cheap, and expendable. A bipedal humanoid that costs $150,000 and falls over on rough terrain is none of those things. Defence tech venture capital hit a record $49.1 billion in 2025, nearly double the prior year, and Goldman Sachs projects 50,000 to 100,000 humanoid robots shipped globally in 2026 across all sectors. Surging defence stocks that signal huge potential for military tech startups have created a funding environment where the pitch “humanoid robot soldiers” opens cheque books. Whether the technology justifies the pitch is a question the battlefield will answer, and the battlefield, so far, favours wheels over legs.