When Agility Robotics deployed Digit units into Amazon's fulfilment centres in 2023, no regulatory body issued a permit for that deployment. No government agency conducted an independent safety audit before the robots went on shift. No standards body had certified the robot's behaviour around human co-workers against a published humanoid-specific benchmark. The deployment happened under the existing patchwork of occupational safety rules written for other kinds of industrial equipment, extended to a category of machine those rules were never designed to address.
That is the actual regulatory situation humanoid robotics is operating in right now. Not a permissive regime with clear rules. Not a prohibitive one either. Something more awkward: a gap, filled mostly by the companies themselves making their own safety determinations, with minimal external oversight.
The gap is not permanent. Governments in the United States, European Union, Japan, and South Korea are all at various stages of thinking about how to regulate advanced robotics and AI systems. But thinking about it and having functional frameworks in place are different things, and the pace of deployment is moving faster than the pace of rulemaking. Understanding where the regulatory picture actually stands — and where it doesn't — matters for anyone trying to assess the real constraints on how quickly humanoid robots can spread into workplaces, public infrastructure, and homes.
What Existing Rules Actually Cover
The starting point for understanding the regulation gap is knowing what existing frameworks do and do not address.
In the United States, the primary body of law relevant to robots in workplaces is the Occupational Safety and Health Act of 1970, administered by OSHA. OSHA's general duty clause requires employers to provide a workplace free from recognised hazards. More specifically, OSHA's machine guarding standards (29 CFR 1910.212) and its robot safety guidelines — largely derived from the American National Standards Institute (ANSI) and the Robotic Industries Association's RIA R15.06 standard — govern industrial robots in manufacturing settings.
The RIA R15.06 standard is a reasonable framework for conventional industrial robot arms. It addresses risk assessment, safeguarding methods, and the requirements for collaborative robot operation (robots designed to share workspace with humans, known in the industry as "cobots"). But it was written for fixed-installation robots with defined workspaces. Humanoid robots that move through an entire facility on two legs, interact with varied surfaces and objects, and work alongside humans in unstructured settings don't fit cleanly into the framework the standard describes. OSHA has not issued specific guidance on humanoid robots, and the Robotic Industries Association has not yet published a humanoid-specific standard, though discussions are underway.
Product liability law provides a separate layer of accountability. If a humanoid robot causes injury, the manufacturer, and potentially the deploying employer, may be liable under existing product liability doctrine. But product liability is reactive — it provides a mechanism for legal remedy after harm has occurred, not a preventive framework that establishes what a safe system looks like before deployment. The absence of pre-deployment certification requirements for humanoid robots means there is currently no mandatory process for demonstrating that a system is safe enough to put in a workplace in the first place.
The EU's Approach: More Ambitious, Still Incomplete
The European Union has taken a more proactive posture toward AI and robotics regulation than the United States, most visibly through the EU AI Act, which entered into force in August 2024 and is being phased in over a multi-year implementation period.
The AI Act classifies AI systems by risk level and imposes requirements accordingly. Systems classified as "high-risk" — a category that includes AI used in safety-critical applications like industrial machinery and certain infrastructure — face mandatory conformity assessments, technical documentation requirements, human oversight provisions, and registration in an EU database before they can be placed on the market. The Act also introduces rules around transparency and prohibits certain applications entirely.
Whether humanoid robots fall cleanly into the AI Act's high-risk categories depends on how they are used. A humanoid operating in a general warehouse alongside human workers doing repetitive tasks looks like it should qualify as high-risk under the Act's machinery safety provisions. A humanoid used for research or demonstration in a controlled environment probably does not. The practical application of the Act to specific humanoid deployments will depend on interpretation that regulators and courts have not yet fully provided.
Separately, the EU's Machinery Regulation (EU) 2023/1230, which replaced the older Machinery Directive and applies from January 2027, includes provisions specifically addressing autonomous systems and collaborative robots. It requires manufacturers to conduct risk assessments for machines with autonomous functions and to demonstrate that those systems behave safely in the conditions of intended use. Humanoid robots deployed commercially in the EU will need to demonstrate compliance. What that compliance process actually requires in practice for a novel category like full-body autonomous humanoids is still being worked out.
The EU's approach is more structured than the American one, but "more structured" does not mean complete or tested. The AI Act and Machinery Regulation are frameworks that will require extensive secondary guidance, technical standards development, and enforcement practice to operationalise. The European standardisation bodies — CEN and CENELEC — are developing harmonised standards that will define what compliance means in technical terms, but that process takes years.
Japan and South Korea: Different Priorities
Japan and South Korea, both significant humanoid robotics hubs, have approached regulation differently — and in ways that reflect their different industrial contexts.
Japan's government has historically taken a promotional posture toward robotics, viewing the sector as central to its industrial strategy and as a partial response to demographic decline. The Ministry of Economy, Trade and Industry and the Ministry of Health, Labour and Welfare have been involved in developing guidelines for robot safety in workplaces, and Japan has an active domestic standards process through the Japan Industrial Standards framework. But Japan's approach has tended toward guidance and facilitation rather than hard regulatory requirements, with a preference for industry self-governance supplemented by government support.
South Korea has similarly emphasised robotics as a strategic industrial priority, with the Robot Industry Promotion Act providing a framework for developing the sector. South Korean regulators have been considering specific rules for service robots — robots operating in public spaces, hospitals, and retail environments — though humanoid-specific standards remain in development.
Neither country has comprehensive humanoid-specific safety regulations in place for commercial deployment. Both are likely to move incrementally through existing machinery and product safety frameworks rather than creating wholly new regulatory categories, at least in the near term.
The Standards Problem
Regulation and standards are related but distinct. Regulations are legal requirements imposed by governments. Standards are technical documents, developed by standards organisations, that define how to measure compliance with regulatory requirements or best practices. Effective regulation of a new technology typically depends on standards that don't exist yet — and standards development for humanoid robotics is genuinely behind.
ISO, the International Organization for Standardization, has published standards relevant to service robots (ISO 13482) and collaborative robots (ISO/TS 15066). ISO 13482, adopted in 2014, covers personal care robots and addresses safety requirements for robots that physically interact with people. ISO/TS 15066 addresses collaborative robot systems in manufacturing environments. Neither was written with full-body autonomous humanoids in mind.
ISO Technical Committee 299, which covers robotics, has working groups examining how existing standards apply to humanoid and autonomous mobile platforms, and what new standards may be needed. But standards development timelines are measured in years, and published humanoid-specific international standards are at minimum several years away. In the interim, manufacturers are making safety claims based on internal testing and risk assessments conducted against frameworks that were not designed for their systems.
This is not necessarily catastrophic — it is the normal condition for novel technologies entering the market before regulation catches up. But it means that current safety claims about humanoid robots should be understood as manufacturer-assessed rather than independently verified against published benchmarks. That distinction is relevant to anyone evaluating deployment claims.
The Public Space Question
Most current humanoid deployments are in controlled environments: factories, warehouses, research facilities. The regulatory complexity increases substantially when humanoid robots move into less controlled settings — hospitals, retail stores, office buildings, and eventually public streets and transit systems.
Public space operation raises questions that workplace safety frameworks do not address. Who is responsible for a humanoid that collides with a pedestrian on a public pavement? Does a robot operating in a hospital corridor need to be registered or licensed, and with whom? What data does a robot collect from its environment as it navigates public space, and what privacy rules apply? When a humanoid is operating in a customer-facing retail setting, what disclosure is required?
Some of these questions have partial answers in existing law — product liability, data protection regulations like GDPR in Europe, and general negligence doctrine — but none were designed specifically for a robot that moves like a person through shared human environments. The most forward-looking municipal governments have begun thinking about autonomous robots in public spaces, primarily in the context of delivery robots and autonomous vehicles, but humanoid-specific public space rules are essentially nonexistent.
Why the Gap Matters Now
The absence of comprehensive humanoid robot regulation is sometimes framed as an opportunity — a sign that the industry has room to develop without being constrained by premature rules. That framing is not wrong, but it is incomplete.
The gap also means that deployment decisions currently rest almost entirely on the judgement of companies with a financial interest in deploying. Workplace safety assessments for humanoid pilots are typically conducted by the deploying company and the manufacturer together, without independent verification. Incident data, when it exists, is not publicly reported in any systematic way. Workers sharing space with humanoid robots in current pilots generally have no formal mechanism through which their safety concerns are assessed against an independent standard.
None of this means existing deployments are unsafe. The companies involved have genuine reasons to care about safety beyond regulatory obligation: a serious incident would be commercially damaging and would likely accelerate restrictive regulation. But "companies have incentives not to hurt people" is not the same as "there is independent verification that they have not." The difference between those two things is what regulatory frameworks are supposed to provide.
The regulatory picture will develop as deployments scale. How prescriptive those rules become, how well they harmonise internationally, and how quickly enforcement catches up to deployment will shape which applications prove commercially viable and at what pace. Those are policy decisions, not technical ones — and they are being made largely without public debate right now. The next few years of rulemaking, standards development, and enforcement practice will set terms that are difficult to revise once the industry has built around them.