A robot injures a person. Who pays?

When a robot injures a person, the first question is always "who pays?" The answer is never simple, because the liability doesn't sit with one entity. It splits across a chain of parties, each of whom contributed to the system that caused the harm.

The AI vendor. If the AI model produced an output that a reasonable safety analysis would have flagged — and the vendor shipped it anyway — product liability applies. Courts are increasingly treating AI software as a product, not just a service. The "black box" defense — arguing that the AI's decision-making is too opaque to assign fault — is weakening. Tesla's Autopilot lawsuits and the Character.AI chatbot case are establishing precedent: if a model is alleged to cause harm, the operator faces claims for design defect and failure to warn.

The OEM. The original equipment manufacturer integrates the AI into the physical machine. If the machine's design or manufacturing created the hazard, traditional product liability applies: design defect, manufacturing defect, or failure to warn. The OEM inherits risk from every component supplier, including the AI vendor. When a robot arm in a Tesla factory injured an engineer, or when a South Korean industrial robot fatally crushed a worker, the OEM was the primary target of legal scrutiny.

The integrator. The integrator installs, commissions, and configures the system at the customer site. If the installation, parameter settings, or safety function validation was the proximate cause of the injury, liability follows. The integrator sets the boundaries — speed limits, force thresholds, zone definitions. If those parameters were wrong, the integrator bears responsibility even if the underlying machine was correctly designed.

The operator. Even with perfect engineering and correct installation, the operator can cause harm by ignoring safety procedures. If lockout/tagout was incomplete, if a safety function was overridden, or if the operator entered a restricted zone, they bear partial or full liability under negligence doctrine. OSHA's Severe Injury Reports consistently show procedure violations as a contributing factor in robot-related accidents.

The practical question isn't whether liability exists in each of these relationships — it's how cleanly fault can be assigned when multiple parties contributed to the incident. That's where hardware-enforced safety becomes relevant not just as a safety mechanism, but as a liability mechanism. A hardware safety layer that operates independently of the AI software creates a clear, auditable boundary: the AI made a decision, the hardware either permitted or blocked the physical outcome. That boundary makes every arrow on the liability map easier to trace.