Insurance for humanoid robots

What a humanoid changes for an underwriter

A wheeled robot at rest is stable. A biped is a controlled fall that never finishes, and an underwriter who has read one incident report knows that a machine the weight of an adult losing balance next to a person is a bodily injury claim with a hospital bill attached. The first question on every humanoid file is therefore how the machine fails, and whether it can fail onto someone.

The second change is where it works. A fenced industrial arm lives inside a cell that people leave before it moves; a humanoid is built for the aisle, the hallway and the kitchen, which are the places general liability forms assume only people go. Autonomy adds a third change: the behaviour is a learned policy or a teleoperator's hands, and there is no rulebook an underwriter can read to see what the machine will do at the edge of its training.

Fleet counts are small. A humanoid company insures a few units or a few dozen, so there is no loss history to price from and each unit is a large share of the scheduled equipment value. Underwriters price what they cannot see with a margin, and a small fleet with no incident logs is mostly things they cannot see.

Which lines respond to a humanoid

General liability responds to injury and property damage while you operate the unit: a pilot on a customer floor, a demo, a unit your technicians run under a service agreement. Products-completed operations coverage, usually a part of the same policy, takes over once a unit is sold and someone else runs it, because the injury then arises from your product rather than your operation. A company that sells units and also runs pilots needs both parts read, since the two are rated on different bases.

Equipment coverage, written as inland marine, follows each unit in transit and at pilots at its replacement cost. Humanoids ship in crates, sit in customer buildings for months, and get moved by forklifts that have no idea what they are carrying, and none of that is inside a standard property policy. Tech E&O enters when the control stack itself is licensed: a foundation model or a locomotion policy sold to run on someone else's body, where the loss is the customer's money rather than a broken hip.

Workers' compensation covers your own technicians, who spend the pilot standing within arm's reach of the machine. It is required by state law once you have employees, and it is the line site owners check most carefully before letting your people on their floor. An umbrella joins when a hospital or utility asks for $5M and your primary carries $1M.

What humanoid coverage costs

No sourced premium figure exists for humanoid robots, and we do not publish one. The published startup cost surveys describe software companies, and the robotics class is new enough that no trade survey has a humanoid row. Anyone quoting a typical number for this class is guessing.

What moves the price is legible. Where the unit operates sets the general liability rate: a fenced test floor, a warehouse aisle with trained staff, a public-facing retail pilot. The scheduled value of the fleet sets the equipment premium, and a small fleet of high-value units pays a higher rate per dollar than a large fleet of cheap ones. The supervision ratio matters most, because a unit with a technician beside it is a very different risk from one running unattended overnight.

The rest is the file. Incident logs, a written operating envelope, an emergency stop that a person can reach, and pilot agreements that allocate liability clearly all reduce what the underwriter has to price blind. A humanoid company with a clean incident log gets terms a company with no log cannot.

The liability chain and AI exclusions

When a foundation model drives the body, four parties sit in the chain: the manufacturer of the hardware, the developer of the model, the integrator who fitted it to the site, and the deployer running it. A European insurer's magazine piece on humanoid liability (2026-07-08) walks that chain and lands where the contracts do, on the deployer in the first instance, with recourse up the chain depending on what caused the fault. A trade piece in PYMNTS (2026-06-23) put it more bluntly: physical AI is forcing insurers to start from scratch, because the forms were written for either a product or a piece of software, and a humanoid is both at once.

Your customer's contract is with you, and the model provider's terms push liability downstream. If you build the body and license the model, you are two links of the chain and hold most of it. If you integrate someone else's body with someone else's model, the indemnities between the three of you decide who pays, and they should be written before the pilot rather than after the fall.

The forms have started to exclude what drives the machine. The standard-forms bureau's generative AI exclusions for general liability took effect in January 2026 (Independent Agent magazine, 2025-10-21), and large insurers won state approval during 2026 for filings that drop AI-related damages from corporate liability policies (The Information, April–May 2026). A humanoid running a learned policy fits the bureau's definition closely enough to be argued over. Read every quote for the words autonomous, artificial intelligence and machine learning before anything binds.

How to place a humanoid program

Present the operation, in the underwriter's order. State the operating envelope: where the unit works, how fast it moves, what it lifts, what it never does. Describe the emergency stop and who can reach it, the supervision ratio during pilots, and whether the unit is teleoperated, autonomous, or both. Say which of the machinery and collaborative-robot safety standards your safety case follows, attach the incident log even if it is empty, and include the pilot agreement so the underwriter can see who is liable on the customer's floor.

Go to insurers with appetite for robotics. A generalist insurer reading a humanoid through a standard application sees a machine that falls on people and declines, and that decline describes the insurer's appetite, not your company. Specialist markets want the risk described this way, and the same operation that was declined routinely quotes once it reaches them.

Then read for autonomy and AI exclusions. Two quotes at the same premium can behave differently at claim time depending on whether the form excludes losses arising from artificial intelligence, and the certificate will not tell you which one you bought. We place through insurers who cover autonomy explicitly, read the policies ourselves, and run the whole placement so you can stay focused on the machine.

This page describes coverage in general terms. It is not an offer of insurance, and carrier appetite, policy wording, licensing, and availability govern every quote and every claim.