When a forklift injures a worker in a warehouse, the insurance claim process is well-worn. The machine is a known category. Its risk profile has been studied for decades. Actuarial tables exist. Underwriters know roughly what they're pricing when a logistics company adds a fleet of forklifts to its commercial general liability policy.
When a humanoid robot injures a worker, none of that infrastructure exists. The machine is a new category. There are no actuarial tables, no decades of incident data, no established underwriting criteria. The insurer is being asked to price a risk they have never seen before, in a technology they may not fully understand, operating in environments where the human-machine interaction has no historical analogue.
This is not a peripheral problem for the humanoid robotics industry. It sits close to the centre of what commercial deployment actually requires. A manufacturing company that wants to deploy humanoid robots at scale needs to know that its existing commercial insurance policies cover incidents involving those robots — or that it can obtain coverage that does. In many cases right now, the answer to both questions is uncertain, and that uncertainty is slowing down commercial commitments in ways that the industry's own coverage rarely discusses.
How Industrial Insurance Usually Works
Commercial businesses that operate machinery typically carry several overlapping insurance products. Commercial general liability covers bodily injury or property damage caused by the business's operations. Product liability covers harm caused by a product the business makes or sells. Workers' compensation covers employees injured on the job, regardless of fault. Umbrella or excess policies provide additional coverage above the limits of primary policies.
For conventional industrial robots — fixed-arm systems that weld, cut, or assemble in enclosed cells — underwriters have decades of claims experience to work from. They know the common failure modes, the typical injury patterns, the effect of proper guarding and safety protocols on claim frequency and severity. A manufacturer deploying a new welding robot isn't introducing a novel risk category; it's adding instances of a well-understood one.
Humanoid robots break that model in several ways simultaneously. They operate in open environments shared with humans, not enclosed cells. They move through space autonomously using software-driven decisions, not fixed mechanical paths. They are designed to interact physically with objects and, increasingly, with people. And when something goes wrong, the causal chain — from sensor input to software decision to physical action to outcome — is considerably more complex than anything insurers have previously had to reconstruct in a claim investigation.
The Causal Chain Problem
In any insurance claim involving physical harm, the insurer needs to establish cause. Who or what was responsible for the incident? The answer determines which policy responds, how coverage is allocated between parties, and whether any subrogation claims (where an insurer pursues a third party to recover what it paid out) are viable.
For a humanoid robot incident, establishing cause is genuinely hard. The robot's behaviour at any given moment is the product of hardware, firmware, the operating software stack, the training data used to develop the robot's capabilities, the specific deployment configuration set by the operator, the conditions in the physical environment at the time of the incident, and potentially remote updates or changes pushed by the manufacturer after initial deployment. Each of those factors is a potential contributor to causation. Each belongs to a different party.
The manufacturer might argue that the incident resulted from the operator's configuration choices. The operator might argue that the incident reflected a hardware defect or a software failure that was the manufacturer's responsibility. If the robot uses a foundation model (a large AI system trained on broad data and fine-tuned for specific tasks) developed by a third party, there may be a fourth party in the causal chain. Insurance policies are not designed to allocate causation across this kind of distributed technical architecture, and courts haven't yet established precedents that would do so consistently.
Some underwriters describe this as the "black box" problem: the robot makes a decision, something bad happens, and reconstructing exactly why the decision was made requires access to data, logs, and expertise that the insurer may not have and may not be able to compel. For conventional machinery, causation is usually mechanical and visible. For an autonomous robot running learned software, it often isn't.
What Underwriters Are Actually Doing Right Now
The insurance industry's response to humanoid robots so far has been cautious and largely ad hoc. A small number of specialist insurers — mostly in the Lloyd's of London market and a few technology-focused US carriers — are writing coverage for early humanoid deployments, but the terms are being negotiated deal by deal rather than drawn from standardised products.
What those negotiations typically involve: detailed technical disclosures from the robot manufacturer about the system's architecture, safety features, and operational constraints; requirements for specific deployment protocols including human oversight ratios and geofencing (restricting where the robot can operate within a facility); exclusions for autonomous operation above certain capability thresholds; and premium structures that reflect the underwriter's uncertainty rather than actuarial precision. In plain terms: coverage is available for early deployments, but it is expensive relative to conventional industrial machinery, restricted in scope, and conditional on operating parameters that some deployments may find constraining.
Workers' compensation introduces a separate complication. In most US states, workers' compensation is a no-fault system: an injured employee is compensated regardless of whether the employer or a third party was responsible for the incident. But when the proximate cause of a workplace injury is a robot, the workers' compensation insurer has a potential subrogation claim against the robot manufacturer or operator. Pursuing that claim requires establishing that the robot's behaviour constituted a defect or negligence — which loops back to the causal chain problem. Workers' compensation carriers are aware of this and are beginning to ask questions about humanoid robot deployments during underwriting that they weren't asking two years ago.
The Data Gap Underneath Everything
Insurance pricing is built on loss data. Underwriters price risk by looking at how frequently similar risks have produced claims and how large those claims were. For a risk category with no claims history, pricing is largely guesswork with a generous margin for uncertainty — which is another way of saying the coverage is expensive.
The humanoid robotics industry has a strong incentive to generate and share operational safety data that would let underwriters price risk more precisely. More precise pricing means lower premiums, which means lower total cost of deployment, which means the business case for buying humanoid robots improves. But sharing operational data creates competitive risks for the robot companies — incident data reveals information about failure modes, capability limits, and operating conditions that competitors could use. It also creates potential legal exposure if the data shows patterns that could support a negligence claim.
Several industry groups are discussing frameworks for anonymised aggregate safety data sharing, modelled on approaches used in aviation and automotive safety. The Aviation Safety Reporting System, which allows pilots and operators to report safety incidents anonymously without legal penalty, is often cited as a model. Whether humanoid robotics companies will actually participate in something similar — given the competitive and legal stakes — remains to be seen. The conversations are happening. Commitments are not yet forthcoming.
The Product Liability Layer
Separate from the coverage that operators need for their deployments, humanoid robot manufacturers face their own insurance challenge: product liability for the robots they sell or lease.
Product liability coverage protects a manufacturer against claims that harm was caused by a defect in its product. For software-driven autonomous systems, the definition of "defect" is legally unsettled. A physical defect — a component that fails in a way it shouldn't — is relatively straightforward. A behavioural defect — the robot made a decision that a reasonable system should not have made — is much harder to characterise, particularly when the behaviour emerged from learned patterns in training data rather than explicitly programmed rules.
The EU AI Act, which entered full application in 2026, introduces risk-based requirements for AI systems used in certain contexts, including requirements for documentation, transparency, and human oversight that have implications for how product liability would be assessed under European law. In the United States, there is no equivalent federal framework, and the applicable product liability standards vary by state and are being litigated rather than legislated. Robot manufacturers selling into multiple jurisdictions are navigating multiple overlapping and sometimes conflicting legal frameworks, with insurance products that don't always map cleanly onto any of them.
Why This Slows Commercial Deployment
The practical effect of insurance uncertainty on commercial humanoid deployment is straightforward: it adds cost, adds time, and in some cases adds a hard stop to deals that might otherwise proceed.
A procurement team at a large manufacturer that wants to deploy humanoid robots needs to confirm that its existing insurance policies cover the deployment or that it can obtain new coverage. If the answer is "yes, with restrictions," the team needs to understand what those restrictions mean for operational plans. If the answer is "we need to get a bespoke endorsement and it will take six to eight weeks to negotiate," that's a delay that adds friction to a procurement process that already has plenty of friction. If the answer is "your current policy explicitly excludes autonomous AI systems," the team has a gap to fill before the deployment can proceed.
None of these outcomes is fatal to the industry. Insurance markets have adapted to new technology categories before — aviation, nuclear power, cyber risk, autonomous vehicles — and humanoid robots will eventually have standard products written for them. But the adaptation takes time, and in the current window, the absence of clear, affordable, standardised coverage is a real constraint on the pace of commercial rollout. It doesn't show up in demo videos or funding announcements. It shows up in the gap between signed partnerships and actual deployed units, which for every humanoid robot company right now is wider than their public communications tend to suggest.
The companies that engage seriously with this problem — building relationships with specialist underwriters, investing in safety data collection and transparency, designing hardware and software with insurability in mind rather than treating it as someone else's problem — will have a structural advantage as the market matures. Insurance is not the most exciting part of the humanoid robotics story. It may, however, be one of the most consequential.