Japan builds more of the world's industrial robots than any other country and runs one of the largest operating fleets, which means the governance of AI inside factories and robotic systems is not a hypothetical for Japanese manufacturers. It is already live on the shop floor, layered across decades of industrial safety law, a fast-evolving voluntary AI framework, and a new national industrial policy that is betting hundreds of billions of yen on AI-driven robotics. Understanding how these pieces fit together, and which ones actually bind a manufacturer, is the practical starting point for governance in this sector.

Japan's manufacturing and robotics base

Japan is the world's second-largest market for industrial robots by annual installations and holds one of the largest operational stocks of installed units, according to the International Federation of Robotics' World Robotics 2025 report, with roughly 44,500 units installed in 2024 and an operational stock of around 450,500 units. Automotive and electronics manufacturing remain the two largest buyer industries. This installed base, much of it running programmable but not learning-capable robots, is the substrate onto which AI capabilities, machine vision for quality inspection, predictive maintenance, adaptive motion planning, and increasingly generative and multimodal models, are now being layered.

That layering is now a matter of explicit national policy. Around 30 June 2026, Japan's Ministry of Economy, Trade and Industry, working with its innovation agency NEDO, selected the SoftBank, NEC, Sony, and Honda-backed venture Noetra alongside the national research institute AIST to lead a five-year, multibillion-dollar Multimodal Foundation Model Development Project for AI Robots and Physical AI, with the program reported to begin operations on 1 July 2026. The project is intended to build a domestic "physical AI" foundation model that gives robots the ability to understand and act in unstructured real-world environments. METI has paired this with a revised national robotics strategy targeting roughly 10 million AI-equipped robots deployed across more than a dozen sectors, including food manufacturing, by 2040. This sits on top of Japan's longer-running public-private robotics coordination efforts, including the 2015 New Robot Strategy, compiled by the Robot Revolution Realization Council and formally decided by the Headquarters for Japan's Economic Revitalization, a Cabinet body, aimed at keeping Japan at the center of robot innovation and adoption.

The structural point: safety law binds, AI law guides

As with other sectors, the governing reality in Japanese manufacturing is that Japan's AI-specific instruments, the AI Promotion Act, the AI Guidelines for Business, and the AI Basic Plan, are voluntary or framework-level, while what actually binds a manufacturer deploying AI in production comes from law that predates AI entirely: the Industrial Safety and Health Act and its implementing ordinance, the Product Liability Act, and, where worker or shop-floor personal data is involved, the Act on the Protection of Personal Information. Manufacturers that treat the AI-specific layer as the compliance obligation and the safety and liability layer as background risk have the analysis backward.

Occupational safety regulation of industrial robots

Japan has regulated industrial robots as a distinct category of workplace hazard since the early 1980s, under the Industrial Safety and Health Act (Act No. 57 of 1972) and its implementing Ordinance on Industrial Safety and Health, both administered by the Ministry of Health, Labour and Welfare (MHLW). The Ordinance contains a dedicated section on industrial robots, Articles 150-3 through 151, which historically required that an operating industrial robot be enclosed by a safety fence to keep workers out of its operating envelope. On 24 December 2013, MHLW issued a notification amending the implementation directive (tsuutatsu) for Ordinance Article 150-4, clarifying that collaborative operation, where a robot and a human work in the same space without a fence, is permitted, aligning Japanese practice with the direction international standards were already taking, provided the risks are assessed and appropriate safeguards, such as power and force limiting or speed and separation monitoring, are put in place.

Technical conformity is anchored in Japanese Industrial Standards that mirror the relevant ISO robot safety standards: JIS B 8433-1 and B 8433-2 correspond to ISO 10218-1 and ISO 10218-2 (robots and robot systems), and TS B 0033 corresponds to ISO/TS 15066, the technical specification governing collaborative robot operation. The Japan Industrial Safety and Health Association (JISHA), established in 1964 under the Industrial Accidents Prevention Organization Act, and the Japan National Institute of Occupational Safety and Health (JNIOSH) support this regime with technical guidance, risk assessment training, and applied research on human-robot collaboration, though the binding requirements sit in the ISHL and its Ordinance, not in JISHA's own materials. A manufacturer introducing an AI-driven robot, for example one whose motion planning or task sequencing is generated or adapted by a machine learning model rather than fixed programming, still has to satisfy this pre-existing framework: a risk assessment covering the AI-influenced behavior, appropriate physical or logical safeguards, and conformity with the applicable JIS standards before the system operates around workers. Nothing in Japan's AI-specific legislation supersedes or substitutes for this.

How the AI Promotion Act and the AI Basic Plan apply

The Act on Promotion of Research and Development, and Utilization of AI-related Technology (Act No. 53 of 2025), Japan's first national AI law, passed by the Diet on 28 May 2025 and promulgated on 4 June 2025, applies to manufacturing exactly as it applies elsewhere: as a promotion and coordination law rather than a regulatory one. It creates no manufacturing-specific obligations, imposes no penalties, and requires no conformity assessment for AI-driven production or quality systems. Its only business-facing duty, the Article 7 duty to endeavour to cooperate with government AI measures, is explicitly non-binding.

What the Act does for this sector is institutional and strategic rather than compliance-generating. It established the Cabinet-level AI Strategy Headquarters, chaired by the Prime Minister, whose provisions took effect on 1 September 2025, and which is tasked with drawing up Japan's AI Basic Plan. The Cabinet decided the first AI Basic Plan on 23 December 2025 and a revised, second plan on 14 July 2026. Both plans name manufacturing explicitly as one of the fields where Japan holds structural strengths and where AI adoption is treated as a national priority, alongside infrastructure, finance, and other sectors. The Basic Plan sets mid- and long-term direction; it is not a source of enforceable manufacturing obligations, and the METI-NEDO physical AI foundation model program described above is best understood as an implementing initiative under this strategic umbrella rather than as regulation.

The AI Guidelines for Business

The joint METI and Ministry of Internal Affairs and Communications AI Guidelines for Business is the instrument most directly relevant to day-to-day governance of AI in manufacturing, even though it remains legally voluntary. Version 1.0 was published on 19 April 2024, consolidating three earlier development, use, and governance guidelines, and set out risk-based principles, including safety, human oversight, and transparency, that apply across AI use cases generally rather than in a manufacturing-specific chapter. The current version, v1.2, was published on 31 March 2026 and added risk framing for AI agents, systems that plan and take multi-step autonomous or semi-autonomous actions, rather than introducing a dedicated Physical AI section. For manufacturers, the practical relevance of this update lies in how the agent risk framing extends to AI systems that plan or sequence actions for robots and production equipment, an increasingly common architecture as generative and multimodal models are layered onto industrial robotics, even though the guidelines do not single out manufacturing, robotics, or logistics as named domains.

Because the guidelines are voluntary, a manufacturer cannot be penalized for failing to meet them as such. In practice they function as the de facto standard of care that regulators, customers, insurers, and courts are likely to reference when assessing whether an AI-equipped production system was reasonably safe, and manufacturers should read the guidelines' safety, human oversight, and risk-management principles alongside the existing safety-fence and risk-assessment logic of the Industrial Safety and Health Act rather than treating the two as separate tracks.

Product liability for AI-embedded equipment

Where an AI capability is embedded in physical equipment, the Product Liability Act applies as it would to any manufactured product. Article 3 makes a manufacturer liable for injury, death, or property damage caused by a defect, defined as a lack of the safety the product should ordinarily provide, in a product it manufactured or processed. AI software in isolation does not constitute a "product" under the Act, but once it is installed in a robot, inspection system, or piece of production equipment, the equipment as a whole, AI included, is a product, and a defect traceable to the AI's behavior can trigger manufacturer liability. Article 4 provides a development risk defense, exempting a manufacturer that proves the defect could not have been detected given the state of scientific and technical knowledge at the time of delivery, but the defense is generally understood to be interpreted narrowly, assessed against the highest level of scientific and technical knowledge available anywhere at the time, which makes it a demanding standard to satisfy in practice. Manufacturers should not treat the development risk defense as a reliable shield; it functions in practice as a high evidentiary bar rather than a routine exemption. There is also no settled Japanese legal standard for what constitutes a "defect" in an AI system whose decisions may not be fully explainable, which is an unresolved area worth watching rather than a solved one.

Data governance on the shop floor

Manufacturing AI, particularly vision-based quality inspection and predictive maintenance systems, runs on shop-floor data that can include images or video of workers, badge and access logs, and operator performance metrics. Where that data identifies individuals, the Act on the Protection of Personal Information applies in the ordinary course, enforced by the Personal Information Protection Commission, which has separately flagged risks specific to generative AI use. Training data drawn from production lines also raises trade secret and competitive sensitivity questions that sit outside AI-specific law entirely, in Japan's Unfair Competition Prevention Act framework for trade secrets, and manufacturers building or fine-tuning models on proprietary process data, including those participating in the Noetra or AIST physical AI initiatives, should treat data provenance and confidentiality as a governance question independent of AI regulation.

Practical governance considerations

  • Run the ISHL risk assessment first. Before deploying an AI-driven robot or AI-adapted process, complete a risk assessment under the Industrial Safety and Health Act and confirm conformity with JIS B 8433 (ISO 10218) and, for human-robot collaboration, TS B 0033 (ISO/TS 15066). This obligation exists independent of anything in the AI Promotion Act.
  • Build in human oversight and emergency stop capability by design. Align physical AI deployments, robotics, autonomous material handling, adaptive production lines, with the safety and human oversight principles in the AI Guidelines for Business, and apply the v1.2 AI agent risk framing where the AI plans or sequences multi-step actions rather than following fixed programming, since these are the benchmark most likely to be applied when assessing reasonable care after an incident.
  • Document the state of the art at deployment. Given the narrow interpretation of the Product Liability Act's development risk defense, keep contemporaneous records of validation testing, known limitations, and the technical knowledge available at the time of deployment for any AI-embedded equipment.
  • Separate data governance from AI governance. Apply APPI requirements to any shop-floor data touching identifiable workers, and treat proprietary process and training data as a trade secret protection question, not merely an AI compliance one.
  • Track the AI Guidelines for Business as a moving target. The guidelines moved from v1.0 to v1.2 in under two years, with the AI agent risk framing added only in the most recent revision; manufacturers should monitor METI and MIC publications rather than treating any single version as settled.
  • Distinguish national industrial policy from regulation. Government-backed initiatives such as the METI-NEDO physical AI foundation model program signal strategic direction and funding priorities, not compliance obligations; participation or reliance on nationally developed foundation models does not substitute for a manufacturer's own safety and liability governance.

The bottom line

Japan's advantage in industrial robotics is real and is now the object of a substantial national AI investment. But the governance question for any manufacturer deploying AI in production, quality control, or robotics is not primarily an AI-law question. It is answered first by the Industrial Safety and Health Act's decades-old robot safety regime and the Product Liability Act's defect liability standard, and only secondarily, as a matter of best practice and reputational and evidentiary risk management, by the voluntary AI Guidelines for Business. The AI Promotion Act and AI Basic Plan set the strategic frame and fund the technology; they do not set the compliance bar.

Primary sources

Related articles