Does the EU AI Act apply to robots?

Yes, through the safety-component rule rather than a robot chapter. An AI system that is a safety component of a product covered by the Annex I harmonisation legislation (the machinery rules are entry 1) is high-risk under Article 6(1), applying from 2 August 2027. A robot that talks to people separately carries Article 50 transparency duties, in application since 2 August 2026.

What is the difference between the Machinery Regulation and the AI Act for a robot?

The Machinery Regulation governs the machine: its physical hazards, its risk assessment and its CE marking. The AI Act governs the AI inside it: when the AI is a safety component, it is a high-risk AI system with its own file. One machine can need both, and the two files were written to reference each other.

Which standard covers collaborative robots?

ISO 10218 parts 1 and 2 cover the robot and the robot system, and ISO/TS 15066 adds the collaborative operation guidance: power and force limiting, speed and separation monitoring. Together they are the usual evidence path for a cobot cell, with ISO 13849-1 or IEC 62061 for the safety-related control system.

How do you certify a robot whose behavior is learned?

Not once and for all. A learned behavior is assured continuously: an evaluation harness before release, a runtime monitor in operation, a safety case that argues the residual risk explicitly (UL 4600 is the template the field converges on), and a change process that re-runs the evidence every time the policy updates. The EU AI Act's high-risk file and the Machinery Regulation's risk assessment both bend toward that model.

Free instrument

Which regime governs your robot?

Physical AI stacks three laws and a shelf of standards: the machine, the AI inside it, the conversation it has with people, and the evidence that holds the whole thing up. Three answers map which layers are yours.

Which laws apply to a robot with AI inside it?

In the EU, the machine itself falls under the Machinery Regulation; an AI system that acts as a safety component of that machine is high risk under Article 6(1) of the AI Act; and a robot that talks to people, or generates content, carries the Article 50 transparency duties. The standards shelf (ISO 10218, ISO 15066, UL 4600) is where the evidence for all three is kept. The navigator below tells you which layers are yours and the application date of each.

1. What kind of machine is it?

2. What does the AI do in it?

The EU AI Act's high-risk tier turns on the last one: an AI safety component of a machine is a high-risk AI system from 2 August 2027.

3. Does it talk to people, or speak with a synthetic voice or face?

The four layers, from the texts

The machine: EU Machinery Regulation (EU) 2023/1230

Risk assessment, technical file, declaration of conformity and CE marking for the machine itself. Applies from 14 January 2027, replacing Directive 2006/42/EC. Written to be coherent with the EU AI Act, so the risk assessment is where learned behavior gets evidenced.

The AI: EU AI Act high-risk tier, Article 6(1)

An AI system that is a safety component of a product under the Annex I harmonisation legislation (machinery is entry 1) is high-risk: risk management, data governance, documentation, logging, human oversight, conformity assessment and registration. Applies from 2 August 2027.

The conversation: EU AI Act Article 50

A robot that interacts directly with people must say it is an AI, and a synthetic voice or face it produces must be marked machine readable and detectable. In application since 2 August 2026, with fines up to EUR 15,000,000 or 3 percent of turnover.

The evidence: the standards layer

ISO 10218 for industrial robots, ISO/TS 15066 for collaborative operation, ISO 3691-4 for driverless trucks, ISO 13482 for care robots, ISO 21448 and UL 4600 for autonomous behavior. Voluntary, but the presumption of conformity runs through them, and a learned behavior without a safety case is an argument you have already lost.

Why this map exists

The Physical AI Hiring Index measures what embodied-AI employers ask for, weekly, from their own job boards. Read the postings and one pattern repeats: safety cases, standards evidence and AI Act readiness named in the same breath as ROS 2 and reinforcement learning. The companies building these machines need people who can hold both files, and the machines themselves need both files held. This navigator is the map of what those files contain.

The transparency duties are quoted in full in the EU AI Act Explorer, Article 50, and the AI-content half of this stack has its own instrument: the AI Content Labeling Checker. Study aid, not legal advice; the official texts are authoritative.

Questions

Does the EU AI Act apply to robots?: Yes, through the safety-component rule rather than a robot chapter. An AI system that is a safety component of a product covered by the Annex I harmonisation legislation (the machinery rules are entry 1) is high-risk under Article 6(1), applying from 2 August 2027. A robot that talks to people separately carries Article 50 transparency duties, in application since 2 August 2026.
What is the difference between the Machinery Regulation and the AI Act for a robot?: The Machinery Regulation governs the machine: its physical hazards, its risk assessment and its CE marking. The AI Act governs the AI inside it: when the AI is a safety component, it is a high-risk AI system with its own file. One machine can need both, and the two files were written to reference each other.
Which standard covers collaborative robots?: ISO 10218 parts 1 and 2 cover the robot and the robot system, and ISO/TS 15066 adds the collaborative operation guidance: power and force limiting, speed and separation monitoring. Together they are the usual evidence path for a cobot cell, with ISO 13849-1 or IEC 62061 for the safety-related control system.
How do you certify a robot whose behavior is learned?: Not once and for all. A learned behavior is assured continuously: an evaluation harness before release, a runtime monitor in operation, a safety case that argues the residual risk explicitly (UL 4600 is the template the field converges on), and a change process that re-runs the evidence every time the policy updates. The EU AI Act's high-risk file and the Machinery Regulation's risk assessment both bend toward that model.

GAGE briefings tell you which AI regulation deadlines are coming, what they actually require of you, and when a program opens.