Robots at the US Border: What Is Actually Deployed

Border Patrol Robots: What DHS Has Agreed To, and What It Has Not

DHS has not agreed to anything with Foundation. A DHS spokesperson told Fox News Digital that "The Department of Homeland Security currently does not have any non-contractual agreements, pilots, or active prime contracts with Foundation." That statement covers all three of the stages people usually confuse: informal arrangements, test pilots, and awarded contracts.

What did happen, according to Fox News Digital's August 2026 reporting, is a presentation. Foundation founder and chief executive Sankaet Pathak said the company is "in conversations with almost all national security government bodies, so Air Force, Navy, Army, DHS," and added: "We continue to have dialogues. We're starting to negotiate contracts and some applications for them." Negotiating is not awarding. No award to Foundation by DHS has been announced.

Pathak's framing of the use case was blunt. "If drones and watchtowers were able to solve for everything, why do we still have humans at the border?" he asked, arguing that "humanoids will fill that gap quite well." He also described a detect-and-flag role rather than an enforcement one: "They will detect and then flag and then a human would decide kind of like what the next step it needs to be." On timing he said, "Tomorrow I think we could do a pilot. Next couple of months we can do something much larger." Those are the vendor's words about the vendor's readiness, not a government schedule.

Treat every claim on this topic by its stage. A demo is a demo. A pilot is a funded trial with a start date. A contract has a number, an awarding office and a public record in federal procurement data. As of August 2026, only the first of those three exists in public reporting.

Who Foundation Is and What the Phantom Actually Is

Foundation is a San Francisco humanoid robotics startup whose Phantom robot is marketed for both industrial work and defence, an unusually open dual-use position for the sector. The company grew out of a 2024 acquisition of Boardwalk Robotics and its cycloidal actuator technology, and it leases fleets on service contracts rather than selling units.

Foundation's own product page publishes a short spec sheet for Phantom: 5 feet 11 inches tall, 176 lb, 88.2 lb payload, 1.7 m/s walking speed, 29 degrees of freedom. It does not publish a battery runtime, and it does not publish an ingress protection rating. Both omissions matter for outdoor patrol work, and we come back to them below.

The defence track record cited in the border coverage is the Ukraine deployment. Foundation confirmed in early 2026 that it sent two Phantom Mk-I units to Ukraine for frontline reconnaissance testing, reported as one of the first humanoid deployments to an active combat theatre. Our full breakdown of the hardware, the funding and the leasing model sits in the Foundation Phantom MK1 review.

One more thing separates a reconnaissance test from a border programme: a war zone tolerates loss rates and improvisation that a domestic law-enforcement agency does not. A robot that fails in Ukraine is a research data point. A robot that fails in Arizona is an incident report, a chain-of-custody question and a public-records request.

What Already Operates at the US Border Today

The southern border already runs on fixed towers, aircraft, aerostats and buried sensors, none of which walk. The Government Accountability Office's 2026 review of the Integrated Surveillance Towers programme (GAO-26-108118) describes roughly 830 fielded towers along US borders, with a plan to reach about 2,300 by 2034.

That expansion is funded. Around $1 billion for autonomous towers comes out of the tax and spending law signed in July 2025, inside a broader roughly $6 billion allocation for border surveillance technology. Anduril won a $363 million CBP contract in December 2025 for more than 200 Extended Range Sentry Towers, and Defense Daily reported the company had delivered more than 40 units and was building more than 15 a month.

Alongside the towers, Customs and Border Protection operates Predator B unmanned aircraft, smaller drones, tethered aerostats under the TARS programme at fixed sites, mobile surveillance vehicles, thousands of cameras and thousands of buried ground sensors. The strategic problem the humanoid pitch targets is real: towers see far but cannot move, and drones move but cannot look under a ledge or into a culvert.

There is also a robot precedent, and it is a quadruped rather than a humanoid. In February 2022 the DHS Science and Technology Directorate published details of testing Ghost Robotics quadrupeds with CBP, first on asphalt, grass and hills in Lorton, Virginia, then on hills, ravines and rock in El Paso, Texas. Officials described the machines as a possible force multiplier that would reduce agent exposure to hazards. Public reporting since then has not documented a fielded operational border patrol robot dog programme, which is itself a useful data point about how long the distance from successful trial to deployment can be.

• Integrated Surveillance Towers: about 830 fielded, target about 2,300 by 2034 (GAO-26-108118)
• Anduril Extended Range Sentry Towers: $363m CBP award, December 2025, 200+ units
• Air assets: Predator B, smaller UAS, TARS tethered aerostats at fixed sites
• Ground: thousands of cameras, buried sensors, mobile surveillance trucks
• Legged robots: DHS S&T quadruped testing with Ghost Robotics from 2022, no documented operational fielding

What a Humanoid Adds That Towers, Drones and Border Patrol Robot Dogs Do Not

A bipedal robot's one genuine advantage over a tower, a drone and a quadruped is that it can operate human-built infrastructure and carry human-shaped gear without a redesign. Ladders, gates, vehicle cabs, door handles and handheld sensor kits are all built to the shape of a person. Everything else Foundation describes for the border role, detecting people, mapping terrain and flagging activity to an officer, is something a tower, a drone or a four-legged robot already does or could do.

Legs beat wheels on broken ground, but two legs do not beat four. A quadruped is statically stable at rest, carries payload on a flat back, and gets back up from a fall far more reliably. For a patrol task that is mostly walking and sensing, the quadruped is the better-matched body plan, which is why the 2022 DHS testing used one. Our Ghost Robotics company page and the industrial quadruped robots guide cover that class in detail.

The published specs make the mismatch concrete, and this is the part missing from most coverage of the story. Ghost Robotics publishes an endurance figure for Vision 60 of about 3.15 hours of continuous walking, an IP67 rating and a rated operating range of -40C to 55C. Foundation publishes neither a runtime nor an IP rating for Phantom. Fox News Digital reported that the current Phantom has weather limitations and that a more rugged second-generation model designed to be waterproof and dustproof is expected in the coming months. A robot that is not yet rated for dust and rain is not yet rated for the Sonoran Desert or the Rio Grande Valley.

The comparison that actually decides procurement is not humanoid versus quadruped. It is legged robot versus another tower. A solar-powered tower watches continuously for years with no charging cycle, no recovery crew and no operator. Any walking robot has to justify its per-hour cost against that baseline.

Cost and Practicality Questions That Have No Public Answer

No public figure exists for what a border patrol humanoid would cost per unit or per patrol hour, because Foundation does not sell Phantom and publishes no price. The company leases fleets on service contracts, so the relevant number would be an annual service rate negotiated per deployment, and none has been disclosed for any government customer.

The unanswered operational questions are more useful than a guessed price. Where does the robot charge, on a border with long stretches lacking grid power? Who recovers a unit that falls in a ravine at 2am, and does an agent then walk into the same terrain the robot was meant to keep them out of? What happens to the connectivity link in a canyon, and does the robot keep patrolling or stop when it loses backhaul? Who holds the video, under what retention schedule, and what happens to footage of people who are not crossing anything?

Maintenance is the quiet cost. Legged robots have dozens of actuators under continuous load, and Phantom has 29 degrees of freedom. Sand, thorn, water crossings and 45C daytime heat are hard on sealed joints and on batteries. A fixed tower has almost no moving parts by comparison, which is a large part of why the tower programme scaled and the robot programmes have not.

For readers pricing robots generally rather than government fleets, our humanoid robot cost guide and the security robots guide cover what commercial buyers actually pay and what the patrol-and-report category delivers today.

Oversight, Legal and Civil-Liberties Questions

The documented objections to border robots come from named organisations and focus on who is watched, not on the robots themselves. The Electronic Frontier Foundation's Dave Maass, quoted by Reason in July 2026 on the tower expansion, said: "The strategy they're talking about now is the same strategy they've had since the 1990s, and it's been a failure since the 1990s." EFF's central point in its own July 2026 analysis is that the people most often captured by border surveillance are the residents who live there, not migrants or cartel operatives.

The ACLU's position centres on the border zone itself. The organisation notes that the federal government claims the power to conduct certain warrantless stops within 100 miles of the border, while stressing that Fourth Amendment protections still apply there. Any new sensor platform in that zone inherits that legal argument along with the hardware.

The programme-history objection is separate and factual. Reason's reporting notes that earlier border surveillance efforts, including SBInet, cost roughly $3.5 billion combined before being discontinued. That record is the reason cost-per-detection and independent evaluation questions get asked early about anything new at the border.

Two questions specific to a humanoid have no answer in the current reporting. The first is arming: Foundation builds explicitly for defence, and reporting on Phantom demonstrations has described the platform handling firearms, while Pathak described the border role as detect-and-flag with a human deciding what happens next. Nothing published states an armed border configuration has been proposed to DHS. The second is disclosure: no federal privacy impact assessment or system-of-records notice covering a DHS humanoid patrol robot has been published, which would normally precede a real pilot.

Border Patrol Robots in China Versus the US Proposal

The Chinese border robot story is a named contract for support work at a crossing, not an armed patrol, and it is a different thing again from a US sales pitch. Reporting describes a roughly $37 million UBTECH Robotics contract to station Walker S2 humanoids at the Fangchenggang crossing in Guangxi, where the described tasks are guiding queues, directing vehicles and checking container identification against customs records. We cover exactly what is and is not confirmed there on the Chinese humanoid robot deployment page, which is the right place for that thread rather than this one.

The structural difference is procurement. A Chinese state operator can field a robot on an internal decision. A US agency deploying the same class of machine passes through competitive procurement, privacy review, congressional appropriation and public-records exposure. That is why a Chinese border robot can appear in service months after a demo and a US one cannot.

What to Watch Next on Border Patrol Robots

The first real signal will be a procurement document, not a press interview. Watch for a CBP or DHS solicitation, a Broad Agency Announcement, or an S&T test announcement naming legged robots, and for any award appearing in public federal procurement records. A DHS Science and Technology Directorate test write-up, like the 2022 quadruped article, is the format this agency actually uses when something real starts.

Watch the hardware milestone too. Foundation's second-generation Phantom, described in reporting as waterproof, dustproof and more rugged, has not shipped. Until a published IP rating and a runtime figure exist, no serious outdoor patrol evaluation can be scored against a tower.

Watch the paperwork. A privacy impact assessment, a system-of-records notice or a Congressional briefing on legged robots at the border would each mark a move from pitch to programme. So would a named test corridor and a named sector chief.

Finally, watch the company's other customers. Pathak said Foundation is negotiating with the Air Force, Navy and Army. If any of those turns into a disclosed award, it tells you the platform cleared a federal evaluation somewhere, which is the strongest available proxy for whether a border evaluation would survive contact with real requirements.