Could humanoid robots be heading for the battlefield?

The US military pilot is limited to handling rather than firing weapons, Pathak says, though weaponisation is part of the testing in Ukraine.

The company attracted attention earlier this year after Eric Trump, the US Presidents' son, became an investor and advisor.

Foundation is also an opportunity for Pathak to prove himself - Synapse, the financial services firm he co-founded and led, filed for bankruptcy in 2024.

But are humanoid soldier robots what the military needs, how hard are they to build and what ethical issues do they raise?

The military is clearly interested, says Dean Fankhauser at Robozaps, a humanoid robotics advisory firm that runs a marketplace for commercial systems. He points to a current US Army contest for humanoids that could eventually support soldiers across a wide range of tasks.

It is "completely inevitable" says Fankhauser that a company would see a business opportunity in weaponising the technology.

There are plenty of simpler robots – namely drones and even some ground robot systems – used to carry explosives, missiles and other payloads, with battlefield use especially visible in Ukraine.

Some firms have also been working to weaponise dog-like quadruped robots, though we haven't seen them too much in active warfare yet notes Fankhauser.

But other legged robot companies have drawn a line opposing weaponisation, citing risks of harm and ethical issues.

Pathak disagrees with that, arguing it is dangerous that more firms aren't following Foundation's lead.

Humanoid robot soldiers make sense, he argues, because the world is built for humans. From screwdrivers to weapons, there is no need to reinvent existing tools.

Humans should be "in the loop", approving any use of lethal force before the system can act, Pathak says, though he makes exceptions where firing autonomously might be necessary to avoid a catastrophic outcome and sees scenarios where human authorisation is less critical.

Perhaps the biggest challenge, and one faced by all companies building humanoid robots, is developing artificial intelligence that can operate in the real world and cope with unpredictable and complicated situations.

Phantom is directed by an AI system called Cortex, and a new version is also in development.

The idea is that Phantom is given a goal – such as moving supplies or mapping the inside of a building – based on a task it has been trained specifically to carry out through demonstrations using videos, images and text.

It then navigates its environment using cameras in its helmet that provide 360-degree vision, allowing its AI system to assess the surroundings and adapt its movements.

In Cortex, says Patak, two types of AI models work together.

A "reasoning model" trained on task-specific examples interprets the goal and formulates Phantom's action plan.

A broader "world model", trained on internet videos as well as data gathered from the robot interacting with the physical world – including its "free play" with blocks – predicts how the environment will respond, helping Phantom move safely and execute actions.

Yet not everyone is convinced the humanoid form factor is the most effective.

Other robots, such as quadrupeds, can navigate terrain more quickly and efficiently, says Fankhauser of Robozaps.

He also notes, based on what he has seen in the commercial space, humanoid technology still has a long way to go.

Today's commercial humanoid robots can barely handle warehouse packing let alone open a door, says Fankhauser.

"If there was a war in Taiwan today, the likelihood that China is going to militarise these humanoids and fight effectively is fanciful," he adds.

While Chinese robots have produced some impressive displays, they have taken place in highly controlled or structured environments – the antithesis of real-world warfare. Though Fankhauser adds things might be different in another five or 10 years.

Robert Griffin works on humanoid robots at the non-profit Florida Institute for Human and Machine Cognition, whose activities include military-funded humanoid projects focused on non-combat applications.

One of its spin-out companies was later acquired by Foundation for part of its core technology.

Griffin sees value in humanoids in reducing risk to human soldiers, but also says unpredictable environments remain a major hurdle.

Getting a robot to jump through a window of unknown height, land on uneven ground, and immediately navigate an unfamiliar interior is hard.

"You get an impression of human-level capability by seeing the human form… but [these autonomous systems] don't know how to handle open-ended uncertainty yet," says Griffin.

Human soldiers have easily foiled AI systems by doing what is "out of the ordinary", like somersaulting or putting cardboard boxes over their heads, he adds.

The practical issues also aren't easily solved.

Runtime is a problem "plaguing every humanoid company" Griffin says – locomotion and moving joints are power consuming. Six hours would be "very impressive".

Whether Foundation can build hands capable of manipulating a weapon designed for a person remains open.

"[The company is] setting extremely hard challenges for their engineering team to either meet or fail at," he says.

Meanwhile, ethical concerns loom large.

Lethal autonomous weapons, whatever their form, lower the barrier to warfare, dehumanise conflict and blur accountability, says Nicole van Rooijen, the executive director of Stop Killer Robots, a global coalition of non-governmental organisations.

But she also finds the humanoid form "extra worrying".

Human-like machines may appear familiar and trustworthy as their civilian use grows, increasing the risk people misread danger.

The answer to the current technological arms race, she argues, is international rules to de-escalate it.