Humanoid robot safety and EU compliance: a practical guide

EU rules for humanoid robots: the Machinery Regulation from 20 January 2027, ISO 10218:2025, the AI Act after the 2026 Omnibus, plus a checklist.

Motion team · 2026-05-18

Last updated: 3 November 2026. Not legal advice: this guide summarises the published texts. For a decision about your own site, talk to your safety advisor or a notified body.

A humanoid robot working on an EU factory floor falls under the Machinery Directive until 19 January 2027 and the Machinery Regulation 2023/1230 from 20 January 2027, the robot safety standards ISO 10218-1 and ISO 10218-2 (2025 editions), and the AI Act, whose high-risk rules reach machinery by 2 August 2028. Every site also needs its own risk assessment.

This guide covers each of those rules, what they mean for the plant that runs the robot, and ends with a checklist for reaching and keeping compliance.

Which regulations apply to humanoid robots

Humanoid robots sit awkwardly in the regulatory framework. Traditional industrial robots are bolted to the floor behind safety cages; humanoids are mobile, autonomous and designed to work alongside people. That puts them under several overlapping rules:

• Robot safety standards: ISO 10218-1 and ISO 10218-2 (2025 editions), which now include the collaborative requirements of ISO/TS 15066, plus ISO 12100 for risk assessment and ISO 13849-1 for safety-related control systems
• The EU Machinery Regulation 2023/1230, which replaces the Machinery Directive 2006/42/EC
• The EU AI Act (Regulation 2024/1689), as amended by the Digital Omnibus on AI (Regulation 2026/1744)
• The Product Liability Directive (EU) 2024/2853, which treats software as a product for anything placed on the market after 9 December 2026

| Rule | Status | Applies from | What it means for a deployer | |---|---|---|---| | Machinery Regulation 2023/1230 | In force since 19 July 2023 | 20 January 2027 | CE marking, safety requirements, rules for software and learning safety functions | | Machinery Directive 2006/42/EC | Applies until repealed | Repealed 20 January 2027 | Covers machinery placed on the market before that date | | ISO 10218-1 and -2:2025 | Published 2025 | Now, replacing the 2011 editions | Robot design and robot application safety, including collaboration | | AI Act 2024/1689 (amended by 2026/1744) | In force since 1 August 2024 | High-risk rules: 2 December 2027 (Annex III), 2 August 2028 (AI in regulated products) | For machinery, AI requirements arrive through the Machinery Regulation by 2 August 2028 | | Product Liability Directive 2024/2853 | Adopted 2024 | Products placed on the market after 9 December 2026 | Software counts as a product; no-fault liability for defects |

The rules are still moving. Humanoids do not fit existing categories neatly, and ISO is writing a standard for robots that balance themselves (see below). Treat compliance as an ongoing process.

ISO 10218 and industrial robot safety

ISO 10218 is the main international standard for industrial robot safety. It has two parts, both revised in 2025 for the first time since 2011:

• ISO 10218-1 sets the safety requirements for the robot itself: its design, construction and built-in protective measures.
• ISO 10218-2 covers robot applications and their integration into the workplace, including cell layout, safeguarding and risk assessment.

The 2025 edition treats collaboration as a property of the application, not of the robot, so there is no longer a separate "collaborative robot" type. It also folds in the requirements of ISO/TS 15066, the 2016 technical specification that used to govern collaborative operation.

ISO 10218-2:2025 recognises these ways for people and robots to share a space:

1. Monitored standstill (formerly safety-rated monitored stop): the robot holds still, under a monitored safety function, while a person is in the shared space.
2. Hand guiding: a person physically guides the robot through its movements.
3. Speed and separation monitoring: the robot slows down or stops depending on how close people are.
4. Power and force limiting: the robot is designed so that any contact with a person stays below pain and injury thresholds.

Humanoid deployments typically rely on speed and separation monitoring, with power and force limiting where contact is possible. The right approach depends on the task, the robot's payload and the workspace layout, and it has to be validated through testing.

Three other standards come up in almost every humanoid deployment:

• ISO 12100 sets the general method for risk assessment and risk reduction of machinery. The site risk assessment below follows it.
• ISO 13849-1:2023 covers the safety-related parts of control systems and the performance level each safety function must reach.
• ISO 3691-4:2023 covers driverless industrial trucks and mobile platforms. It explicitly excludes robots that need active control to stay upright, so it does not cover a robot balancing on two legs.

That last point is the real gap. ISO 10218 does not address the risks of mobility, and a robot that balances on two legs can fall if it loses power. ISO 25785-1, a safety standard for industrial mobile robots with actively controlled stability, is being written for this case. As of November 2026 it is still a committee draft (ISO/CD 25785-1) and has not been published.

EU Machinery Regulation 2023/1230 in brief

The Machinery Regulation replaces the Machinery Directive 2006/42/EC. It entered into force on 19 July 2023 and applies from 20 January 2027, when the Directive is repealed. There is no choice between the two: until that date the Directive applies, and from that date the Regulation applies. Machinery placed on the market before 20 January 2027 stays under the Directive and does not need a new CE mark.

The changes that matter most for humanoids:

• Instructions and technical documentation can be digital.
• Control systems must resist foreseeable malicious attempts that could lead to a hazard, and keep a log of safety software versions.
• A substantial modification, physical or digital, can make the person who made it the manufacturer, with a new conformity assessment.
• Machines whose behaviour evolves must stay safe while they learn. Where a safety function relies on machine learning, the Regulation (Annex I, Part A) removes self-certification: a notified body has to assess it. A robot that learns a task while a separate, conventional safety system handles slowing and stopping is a different case. Ask the supplier which applies, in writing.

The robot maker is normally responsible for CE marking the robot. Whoever assembles the robot into a wider installation, or substantially modifies it, takes on the manufacturer's duties for that whole. The employer's duties for safe use come from the Work Equipment Directive 2009/104/EC.

For dates, substantial modification and the questions to ask your suppliers, see Machinery Regulation 2023/1230: what changes for robot users.

What the EU AI Act means for humanoid robots

The EU AI Act (Regulation 2024/1689) is the first comprehensive law on artificial intelligence. It entered into force on 1 August 2024. The Digital Omnibus on AI (Regulation (EU) 2026/1744, adopted 8 July 2026, in force since 27 July 2026) moved the high-risk dates: 2 December 2027 for the stand-alone high-risk systems listed in Annex III, and 2 August 2028 for high-risk AI built into products covered by EU product law (Annex I).

The AI Act matters for humanoids because AI controls them, from perception and navigation to task planning. Whether a given AI system is high-risk depends on whether it ensures a safety function. Perception or task AI that sits next to a conventional safety controller is not automatically high-risk.

For machinery, the Omnibus moved the Machinery Regulation from Section A to Section B of the AI Act's Annex I. The AI Act's high-risk requirements will reach machinery through delegated acts under the Machinery Regulation, applying by 2 August 2028. In practice, an AI system that ensures a safety function on a humanoid should expect requirements like these:

• Risk management: a continuous risk management process throughout the AI system's lifecycle.
• Data governance: training, validation and testing datasets must meet quality criteria.
• Technical documentation: detailed records of the AI system's design, development and intended purpose.
• Record-keeping: automatic logging of events during operation so they can be traced.
• Transparency: clear information to deployers about the AI system's capabilities, limitations and intended use.
• Human oversight: the system must allow effective human oversight, including the ability to intervene and override.
• Accuracy, robustness and cybersecurity: the system must reach appropriate accuracy and withstand errors, faults and attacks.

Plants deploying humanoids need visibility into the AI running on them. Ask the robot supplier and the deployment partner which functions use AI, which of them are safety functions, and how logging and oversight work.

Product liability: software counts from December 2026

The Product Liability Directive (EU) 2024/2853 replaces the 1985 directive for products placed on the market after 9 December 2026. Three points matter for humanoids:

• Software, including AI, counts as a product, so a faulty software update can create liability in the same way as a faulty part.
• Liability does not depend on fault: an injured person has to show a defect, the damage and the link between them.
• Where the technology is too complex to prove a defect, courts can presume one.

For the injured party this makes claims simpler. Between the robot maker, the software provider and the operator, cause still decides who pays, which is why insurance and clear contracts matter. See insuring humanoid robots.

Risk assessment per deployment

The deployment-specific risk assessment, done following ISO 12100, is one of the most important compliance activities. The robot maker performs a general risk assessment; this one evaluates the hazards in your facility, with your workers, performing your tasks.

It should cover five groups of hazards.

Physical hazards: what contact scenarios can occur between the robot and workers? What forces and speeds are involved? Are there pinch points, crush zones, or places where a worker could be trapped? What happens if the robot falls?

Environmental hazards: what are the floor conditions? Are there slopes, obstacles or wet surfaces? What is the lighting like? Could temperature extremes affect robot performance?

Task-specific hazards: what objects does the robot handle? Are they sharp, hot, heavy or fragile? What happens if the robot drops one? What are the consequences of a task done incorrectly?

Interaction hazards: how do workers interact with the robot? Are there shared pathways or shared workstations? Are there handover points where a person receives an object from the robot?

Cybersecurity hazards: what happens if the robot's network connection drops? What if an unauthorised user gets into the fleet management system? What if a software update introduces a bug in safety-critical behaviour?

The output is a documented list of hazards, risk ratings and mitigation measures. Review it whenever the deployment changes: new tasks, new workspace layouts, new robot models or significant software updates. To prepare the site before the assessment, use the factory readiness checklist.

Safety built into the deployment

How the robot is deployed and run has a large effect on safety. Whoever runs the deployment should be able to show you these controls.

Hard limits on robot behaviour, such as maximum speeds, force limits and permitted work zones, that task instructions and AI learning cannot override. These limits are set per deployment and per robot model.

New tasks are checked against safety rules before they reach the robot, and tasks that break a safety constraint are rejected with a clear explanation.

Every instruction sent to every robot is logged with its source (human operator, workflow, automated schedule) and its outcome. This audit trail provides the traceability both the Machinery Regulation and the AI Act require.

When a robot meets an unexpected situation, it defaults to safe behaviour: stopping, moving to a safe position, or asking for a person to step in. It should never guess its way through an ambiguous safety situation.

Not every user can change safety parameters. Role-based access separates operational tasks (assigning work, scheduling) from safety configuration (speed limits, force thresholds, zone boundaries).

Every Motion deployment includes a risk assessment for your site. Your first humanoid deployment, week by week shows where it sits in the 12 to 15 weeks to autonomy.

A practical compliance checklist

For manufacturers preparing to deploy humanoid robots in a European plant:

Before procurement:

• [ ] Confirm the robot has CE marking under the Machinery Directive, or the Machinery Regulation from 20 January 2027
• [ ] Request the EU declaration of conformity and technical documentation from the robot maker
• [ ] Check which collaborative methods the robot supports under ISO 10218-2:2025
• [ ] Ask which safety functions, if any, rely on machine learning, and whether a notified body has assessed them
• [ ] Understand which AI systems run on the robot and whether any ensures a safety function
• [ ] Check the deployment's safety controls (behaviour limits, audit trails, access control)

Before deployment:

• [ ] Conduct a deployment-specific risk assessment (ISO 12100) covering physical, environmental, task-specific, interaction and cybersecurity hazards
• [ ] Design the workspace layout with safety zones, emergency stops and clear signage
• [ ] Configure robot behaviour limits (speed, force, work zones) for the deployment
• [ ] Train operators on normal operation, emergency procedures and safety protocols
• [ ] Set up monitoring and logging for traceability

During operation:

• [ ] Monitor robot behaviour and fleet performance continuously
• [ ] Review and update the risk assessment when tasks, layouts or software change
• [ ] Keep incident logs and near-miss records
• [ ] Check software updates for safety impact, and for substantial modification, before they go live
• [ ] Run periodic safety audits with qualified people

Ongoing:

• [ ] Track regulatory developments: the Machinery Regulation delegated acts on AI, the AI Act dates of 2 December 2027 and 2 August 2028, and ISO 25785-1
• [ ] Review the robot maker's security advisories and apply patches promptly
• [ ] Document all compliance activities for inspection

If you have a task in mind, book a 30-minute call and we will tell you whether a humanoid fits it and what the safety case would involve.

← Back to blog