Humanoid Robot Failures Put Factory Liability on Multiple Parties

By Amara Osei

Agility Robotics and FORT Robotics announced a partnership in October to strengthen the safety architecture of Agility’s Digit 5 humanoid robot. Unite.AI reports that the plan combines onboard safety systems with external interfaces, communications and manual override. Unite.AAI calls this a significant shift because robots must be safe in human workplaces, which requires controls beyond the robot itself.

A robot can pass a performance test and still become a safety hazard when it encounters an unexpected obstacle, a software update changes its behavior or a worker enters its path, Unite.AI says. The opinion argues that task performance no longer settles the deployment question; factories also need safe operation and a clear allocation of responsibility.

What changes when humanoids enter human spaces?

Traditional industrial robots usually operate inside enclosed work cells separated from workers by barriers and safety devices, according to Unite.AI. Mobile humanoids are designed to navigate spaces built for people, switch tasks and adapt to changing conditions, a flexibility that creates false detections, unwanted movement and hazards created by factory-layout changes.

Unite.AI says a software update can alter movement behavior, while a new factory arrangement can expose conditions absent from earlier tests. Dana R. Wolfe, Vice President and Head of Industry at GFT Americas, told Unite AI that deployments need multiple protection layers, including deterministic software guardrails, factory-specific confidence thresholds, sensor fusion and staged validation before use.

Wolfe also said manufacturers cannot eliminate false-positive results. Their task, he said, is to prevent those errors from becoming unsafe physical movements, with confidence thresholds, human authority and hardware interlocks built into the safety architecture.

Do current industrial robot standards cover mobile humanoids?

The International Organization for Standardization (ISO) has published safety requirements for industrial robots, and ISO 10218-2 addresses their integration into complete systems, according to Unite.AI. Those standards provide a foundation but do not cover every application or hazard involving mobile robots working alongside people.

Wolfe told Unite AI that independent testing, scenario-based validation and phased workplace pilots are important as industry practices develop. Safety evaluations must be repeated whenever changes to the AI model, software, movement restrictions, tasks or factory environment could invalidate earlier results. After a major change, a robot safe in one set of conditions may not be safe in another.

What must human oversight actually do?

Unite.AI says monitoring every AI decision is impractical, and excluding people from critical decisions makes accountability harder to prove. Adrian Figueroa Ramos, co-founder and CEO of Flexio and Routina, said oversight should focus on decisions carrying the most uncertainty or risk.

Ramos told Unite AI: “In today’s AI world, the volume of outputs is more than a human can process. Reviewing everything makes the human the bottleneck, but human judgment is still necessary for high-risk or low-confidence decisions.”

For physical AI, Unite.AI says oversight requires named individuals with authority to override the system. Those overseers must ensure decisions are documented and checked, rather than merely clicking an approval button.

Who is liable when a robot causes harm?

Unite.AI says technical safeguards alone cannot resolve accountability when a robot errs because of a faulty component misidentified by an AI system. The manufacturer, software developer, system integrator and employer could then face liability, with responsibility allocated according to each party’s control over the system.

Under Unite.AI’s framework, manufacturers create and document safety measures. Software developers test changes that could affect perception or movement, while integrators validate the complete system in its intended environment. Employers train workers, maintain safeguards and establish procedures for stopping unsafe operations.

Unite.AI also treats cybersecurity as part of workplace safety. Operational technology linked to robots can be exposed through sensors, software and connected external systems; a security failure that changes physical behavior creates both a safety problem and an IT problem.

Unite.AI expects humanoid robots to deliver more flexibility and productivity in factories, but says adoption will not rest on performance measures alone. It calls for clear responsibility, independent validation and safeguards maintained throughout a robot’s working life.

For factories evaluating humanoids, the next test is whether manufacturers, developers, integrators and employers can document control over safety-critical functions before deployment. Unite.AI says mistakes can occur, but clear accountability, independent validation and lifecycle safeguards must determine whether errors become worker-safety failures.