EU Regulation
EU AI Act & Robots: What It Means for Physical AI Hardware
Robots enter the EU AI Act via two distinct legal routes. The Machinery Regulation adds a parallel compliance layer. January 20, 2027 is the hard deadline — no grace period.
Last updated: July 2026
The two legal routes into the AI Act
Safety component of a product under sectoral law
If a robot's AI system functions as a safety component of a product covered by EU sectoral legislation (such as the Machinery Regulation), it is automatically classified as high-risk under AI Act Annex I.
Example: An industrial robot where the AI controls safety-critical motion in a shared workspace. The AI component is a safety component of the machinery product.
High-risk use case regardless of product type
If the robot performs functions listed in Annex III, it is high-risk regardless of whether it is a product covered by sectoral legislation. Use cases include: biometric identification, critical infrastructure management, employment and worker management.
Example: A security robot that uses biometric identification to grant facility access. The use case (biometric ID) triggers Annex III, not the product category.
Machinery Regulation 2023/1230: what it requires
Enforcement date: January 20, 2027. No grace period. Products with ML-based safety functions placed on EU market after this date must comply. Third-party Notified Body assessment required for ML safety functions.
Notified Body assessment
ML-based safety functions require third-party assessment by an EU Notified Body — an accredited independent testing organization. Self-declaration alone is not sufficient for safety-critical AI.
Human supervisory function
Mandatory design requirement: the system must always allow a human to supervise and override automated decisions. Must be documented in technical file.
Digital documentation
Full technical documentation must be maintained digitally for 10 years from market placement date. Must be available to market surveillance authorities on request.
Safety-related software updates
Any update to software affecting safety functions triggers a new conformity assessment. Cannot push safety-relevant updates to deployed machines without re-evaluation.
Consumer vs industrial: what actually applies
Subject to Machinery Regulation. If AI controls safety-critical functions → also high-risk under AI Act Annex I. Requires Notified Body assessment for ML safety components from January 2027.
No binding EU rules as of 2026 if not classified as safety-critical. A companion robot (aibo, Lovot) does not meet Annex I or Annex III thresholds under current interpretation.
Covered by Medical Device Regulation (MDR) as primary law. AI Act applies as overlay. Both compliance frameworks must be satisfied simultaneously.
Biometric identification use cases → Annex III. Real-time remote biometric ID in public spaces is prohibited (with narrow exceptions). Post-hoc biometric ID is regulated but not prohibited.
High-risk compliance: what you must do
If your robot system is classified high-risk under either route, these obligations apply before EU market placement.