In April 2023, a worker at a South Korean distribution facility was reportedly pinned against a conveyor belt by an industrial robotic arm that had misidentified him as a box of produce it was sorting. The worker died. Investigators found that the robot had been operating outside its designated zone and that a sensor designed to detect human presence had either malfunctioned or been improperly configured.
Determining liability in that case took months and involved the robot manufacturer, the company that had integrated the system into the facility, and the facility operator — each pointing, in various directions, at the others. Industrial robots have a decades-long legal history and established frameworks for sorting out those questions, imperfect as they are. Humanoid robots, which are designed to move through the same spaces as people and interact with them directly, will generate similar incidents. The legal frameworks to handle them do not yet exist in any coherent form.
That gap — between where the technology is going and where the law currently stands — is worth understanding before the first significant case tests it in court.
The Existing Legal Framework and Why It Does Not Quite Fit
When a machine injures someone in a workplace setting, liability in the United States and most other developed legal systems typically flows through one of three channels: product liability (the robot was defective as designed or manufactured), negligence (someone failed in a duty of care — the manufacturer, the operator, a maintenance provider), or workers' compensation (which removes the question of fault from workplace injury claims but compensates workers regardless).
Product liability law has handled industrial machinery for decades. If a robot arm has a design flaw that causes it to behave dangerously under foreseeable operating conditions, the manufacturer can be held responsible under strict liability — meaning the injured party does not have to prove negligence, only that the product was defective and caused harm. This is well-established law.
The complication with humanoid robots — and with autonomous systems generally — is the question of what counts as a defect when the behaviour that caused harm emerged from a machine-learning system rather than from a specific, traceable engineering decision. Traditional product liability assumes you can point to something that was designed or built incorrectly. When a neural network, trained on millions of data points, produces a behaviour in a novel situation that no engineer specifically programmed, identifying the defect becomes significantly harder.
Some legal scholars have proposed adapting the "design defect" concept — arguing that if a manufacturer could have made the system safer through different training procedures, different safety constraints, or different testing protocols, then the resulting harm reflects a design choice that should attract liability. Others argue that this framework will be difficult to apply consistently, because the same argument could hold almost any AI system to an impossibly high standard of retrospective safety.
The Autonomy Problem
The deeper legal difficulty with humanoid robots is autonomy — specifically, the fact that as these systems become more capable, they are making more decisions independently, and those decisions are harder to trace back to any specific human choice.
A conventional industrial robot is programmed to follow a defined sequence of movements. If it injures someone, you can usually answer the question "what decision led to this?" and trace that decision to a programmer, an engineer, or an operator who configured the system. The causal chain is relatively clear.
A humanoid robot operating with vision-based perception and a learned behavioural policy — where its responses to situations are determined by parameters adjusted during training rather than by explicit rules — can reach outcomes that were not directly anticipated by any human who worked on the system. The robot perceives a situation, generates a response based on patterns in its training data, and acts. If that action causes harm, the causal chain runs through the training process in ways that may be genuinely difficult to reconstruct after the fact.
This creates what legal theorists have called the "responsibility gap" — a scenario where harm occurs but no single human actor made the specific decision that caused it, and no existing legal concept assigns responsibility cleanly. The gap is not new to autonomous systems; it has been discussed in the context of autonomous vehicles, medical AI, and algorithmic decision-making for years. But humanoid robots, by virtue of operating physically in spaces with people, raise the stakes considerably.
What Operators Are Currently Doing — and What That Implies
In the absence of settled legal frameworks, companies deploying humanoid robots are managing liability risk primarily through contracts. The agreements between robot manufacturers and operators typically include detailed provisions about permitted use cases, required operator training, maintenance obligations, and — critically — indemnification clauses that specify who bears legal exposure if the system causes harm in various circumstances.
These contracts are not public. But their existence and the specificity with which they reportedly address liability scenarios suggest that both manufacturers and operators are aware that the legal landscape is unsettled and are trying to manage that uncertainty privately rather than waiting for it to be resolved publicly.
The practical implication is that many of the most important early liability decisions in humanoid robotics will be made in private settlements rather than in court, which means the public record will develop slowly and the precedents that would normally guide future behaviour will not be visible. This is not unusual for nascent technology categories — the early years of autonomous vehicle incidents produced a similar pattern — but it does mean that the legal framework will lag behind the deployment reality for some time.
The Workplace Safety Dimension
Separate from product liability and negligence, workplace safety regulation introduces its own layer of legal obligation. In the United States, the Occupational Safety and Health Administration (OSHA) sets standards for machinery safety in workplaces and has the authority to investigate incidents and impose penalties.
OSHA's existing standards for industrial robots were written with fixed-installation machines in mind. They focus on guarding, interlocks, and defined safe zones that keep humans physically separated from robot operation. Humanoid robots working alongside people in shared spaces are architecturally incompatible with the "keep humans out of the robot's zone" approach these standards assume.
OSHA has been developing updated guidance for collaborative robot systems — robots designed to work in proximity to people — and some of this is relevant to humanoids. But as of mid-2026, there is no humanoid-specific OSHA standard, and inspectors investigating an incident involving a humanoid robot in a warehouse would be applying frameworks developed for a fundamentally different class of equipment. Legal liability in the wake of such an incident would be contested on that basis.
The European Union has moved further than the United States on AI-specific regulation with the EU AI Act, which entered into force in 2024 and includes provisions specifically addressing high-risk AI systems used in employment contexts. Humanoid robots in workplaces would likely qualify as high-risk systems under this framework, triggering requirements for conformity assessment, transparency documentation, and human oversight mechanisms. Whether these requirements meaningfully alter liability outcomes in practice remains to be tested in actual cases.
Insurance as a Liability Signal
One practical indicator of how the industry is pricing liability uncertainty is the state of the insurance market for humanoid robot deployments. Industrial robot insurance is a mature product — insurers have decades of actuarial data and can price the risk of conventional robotic systems with reasonable confidence. Humanoid robot insurance, by contrast, is a product that specialist underwriters are still developing.
Several specialist insurers have begun offering coverage for autonomous robot deployments, including humanoid systems, but the premiums reflect significant uncertainty rather than confident actuarial modelling. Operators are being asked to provide detailed documentation of their deployment conditions, safety protocols, human oversight procedures, and incident response plans — requirements that are meaningfully more extensive than those for conventional industrial robot coverage.
Insurance markets have historically served as an early-warning system for liability exposure that regulation has not yet addressed. High premiums and extensive documentation requirements signal that the industry is aware the legal framework is incomplete — and that the financial consequences of incidents could be significant and unpredictable. When actuaries are uncertain, that uncertainty has a price, and operators are paying it.
The Question of Worker Rights and Recourse
The liability question is not only about who pays when a humanoid robot causes physical injury. It also encompasses subtler questions about worker rights in environments where humanoid robots are present.
Workers who believe a robotic system in their workplace poses a safety risk have limited formal mechanisms to compel its removal or modification. OSHA provides the right to report hazards and request inspections, and in principle workers can refuse to work in conditions they reasonably believe are imminently dangerous. But "an autonomous robot whose behaviour I find unpredictable" is harder to operationalise as a legal safety claim than a broken guard rail or an exposed electrical conductor.
Labour agreements in unionised workplaces are beginning to address this more directly. Several recent contracts in logistics and manufacturing have included provisions requiring advance notice before autonomous systems are introduced, joint labour-management safety review processes, and in some cases the right to remove a system from operation pending safety review if an incident occurs. These contract provisions are creating a parallel layer of worker protection that does not depend on waiting for regulatory frameworks to catch up — and they may end up being the most practically significant legal developments in this area over the next several years.
The Precedent Problem
Courts generally reason from precedent — applying principles established in earlier cases to new situations. The challenge with humanoid robot liability is that the earlier cases are imperfect analogies at best.
Autonomous vehicle litigation is the closest parallel, and courts have handled a number of cases involving self-driving vehicles over the past decade. The legal patterns that have emerged from those cases — including how software updates affect product liability analysis, and how to allocate responsibility between the vehicle manufacturer and the operator — are relevant but not directly transferable. An autonomous vehicle has a defined operating domain (roads) and a defined task (transportation). A humanoid robot is supposed to operate across a much wider range of environments and tasks, which makes the analysis more complex.
Some legal theorists have proposed that humanoid robots may eventually require an entirely new liability category — something closer to the framework applied to inherently dangerous activities, where the operator bears strict liability regardless of fault, simply because the activity carries elevated risk that cannot be eliminated through reasonable care. This approach would place the financial burden of incidents on the entities that chose to deploy the technology and benefit from it, rather than requiring injured parties to prove fault. It would also create strong incentives for operators to ensure their systems are safe before deploying them, since they would bear the cost of failures regardless.
Whether courts or legislatures move in that direction is genuinely uncertain. What is not uncertain is that the current state — no clear framework, private settlements, regulatory guidance developed for different equipment — is a temporary condition. The decisions made in the next few years, in courtrooms and legislatures and contract negotiations, will shape a legal landscape that will govern the humanoid robotics industry for a long time.