Articles - Insuring physical AI: Robotics reshaping risk and liability

By Sonal Madhok, Technology and Future Liability Analyst, Willis Research Network 

Morgan Stanley estimates the humanoid robotics market could reach $5 trillion by 2050, with nearly 1 billion robots in use. A more detailed discussion of the market outlook, growth drivers, and adoption trends is available in the companion article.

The technology is already being tested in the real world. In June 2026, Waymo recalled 3,871 robotaxis after reports that some vehicles entered closed freeway construction zones and continued driving at speed. While no injuries were reported, the incident highlighted a growing challenge for autonomous systems: software must interpret and respond to unpredictable physical environments, where road layouts, hazards and operating conditions can change rapidly.

The more AI moves from software into machines that interact with the physical world, the nature of risk changes. A software error that might once have resulted in an incorrect recommendation, biased output or data breach can now lead to bodily injury, property damage, operational disruption and complex liability disputes.

Examples illustrate these risks: warehouse robots colliding with workers, autonomous vehicles running red lights and crashing, a delivery drone dropping a package on a pedestrian, or an AI-controlled manufacturing machine sparking a fire. U.S. occupational data found 77 robot-related accidents from 2015–2022 that caused 93 recorded injuries, including finger amputations and fractures. As adoption grows, both frequency and severity of incidents may increase.

Many incidents originate in software failures. OSHA guidance shows accidents often occur during programming, testing or maintenance, when robots may activate unexpectedly due to control or state-management errors. This risk is heightened in testing environments, where safety mechanisms may be temporarily bypassed.

Cyber threats may also have physical consequences. A compromised robot may not simply lose data; it could disrupt operations, damage property or injure people. As robotics becomes more connected, the boundary between cyber, operational and physical risk continues to erode.

For example, in a warehouse accident where a humanoid robot drops a heavy box on a contractor, liability may fall on different actors:

This creates overlapping exposures across product, general, professional liability and cyber lines, complicating accountability and coverage.

Despite different regulatory approaches, the direction of travel is clear. Responsibility is increasingly being shared between manufacturers and the organizations deploying autonomous systems, placing greater emphasis on governance, operational controls and human oversight.

Key lines of coverage may include:

The convergence of risks creates overlaps and potential gaps. In response, insurers are clarifying coverage through exclusions and endorsements, developing robotics-specific products and exploring integrated policies combining multiple coverage areas. For instance, some cyber insurance policies now explicitly mention coverage (or exclusions) for AI-driven incidents like algorithmic errors or malicious use of AI.

Specialty underwriters are exploring coverage for autonomous fleets, combining elements of auto, product and cyber liability. Reinsurers are also examining systemic loss scenarios, such as software faults affecting large numbers of robots simultaneously. For example, a software update, cloud outage, navigation error or vulnerability in a widely deployed robotic platform could affect thousands of devices simultaneously. This raises accumulation concerns for insurers and reinsurers. Multiple organizations may often rely on the same operating systems, AI models or software suppliers. While such scenarios remain speculative, they underscore the need for agility in insurance.

Innovation is emerging globally. In 2025, China Pacific Insurance launched “Ji Zhi Bao,” a policy covering a robot lifecycle from production to deployment, including property damage, third-party liability and short-term testing risks for humanoid robots. Products such as these illustrate how insurers are moving beyond traditional silos, combining liability and asset protection in a single solution.

Robotics systems depend on specialized inputs such as semiconductors, sensors and batteries, creating exposure to geopolitical and supply chain disruption. Their high value also introduces transportation and logistics exposures, increasing demand for marine, cargo and transit insurance.

Environmental risks also become more important as deployment increases. Battery production, storage and disposal introduce potential pollution, fire and hazardous waste exposures and contamination that may not be fully covered under traditional property policies, increasing demand for specialist environmental liability cover.

As AI moves into these less predictable settings, questions around supervision, maintenance and responsibility for harm become significantly more complex.

This approach is increasingly reflected in international standards governing collaborative and autonomous robots. For example, emerging guidance ISO 25782-1 drafts robot safety for advanced use cases like bipedal humanoids, where they have fail safe mechanisms so a 100kg robot doesn’t topple onto someone during a power loss. Together, these place growing emphasis on human oversight, fail-safe mechanisms, safe human-robot interaction and robust operational controls, particularly as robots move beyond factories into public and domestic environments.

Evidence shows these measures are critical. A study from the New South Wales government on working safely with robots notes that incomplete risk assessments for collaborative robots (cobots) can leave hidden risks that only surface after an injury or near-miss. Risk managers may consider thorough hazard analysis before robots go live, proactively addressing questions like: What happens if the robot loses power or connectivity? How will human supervisors intervene if it malfunctions?

Organizations may consider the following key practices:

These measures reduce operational risk and may improve insurance outcomes. Insurers are increasingly incorporating questions on AI usage and controls in underwriting applications, echoing the way cyber insurers evolved questionnaires on IT security. Demonstrating robust AI governance and safety culture can help companies negotiate better terms or premiums, whereas poor controls might lead to exclusions or higher rates.