Humanoid robots on sale in Spain: the regulatory puzzle surrounding AI
Human-shaped robots are no longer just the stuff of films: in Spain, the CSIC is developing TEO, a humanoid designed to provide care for people, and several hotel chains and airports are already using robots to assist the public. Against this backdrop, Sonke Lund draws on Asimov’s Three Laws of Robotics — do no harm to humans, obey them, and protect the robot’s own existence — to explain how the European AI Act translates this fictional logic into real legal obligations: the more dangerous an AI system is, the more requirements it must meet.
Most humanoid robots will fall into the ‘high-risk’ category, which entails risk management, high-quality training data and high levels of robustness and cybersecurity. The requirement for effective human supervision, ensuring that an operator can understand what the system is doing, intervene and, if necessary, switch it off, reflects Asimov’s Second Law transformed into a design obligation. In practice, before selling a humanoid robot in Spain, it will be necessary to demonstrate compliance, prepare technical documentation, ensure permanent human supervision and register the system in a European database, with the regulations coming into force between 2025 and 2027.
Apart from the AI Act, the author identifies three areas of Spanish legislation that already affect the sale of these robots, although none was designed for that purpose: product safety (originally intended for conventional consumer goods, not for machines that learn and react in unexpected ways); civil liability for defective products, the application of which to AI will raise questions whilst the future European Directive on AI liability is being transposed; and privacy, where the GDPR ties in directly with Asimov’s First Law through data minimisation, purpose limitation and proactive accountability, in an area where the AEPD has already imposed sanctions on technology companies.
The article also addresses the issue of intellectual and industrial property rights relating to these devices: embedded software is protected by copyright, algorithms can be protected as trade secrets, technical inventions are patentable with certain caveats, and the physical appearance of the humanoid can be registered as an industrial design. The most complex question, Lund points out, is what happens if the robot generates something original of its own accord: at present, only human beings can be authors, although technological progress could eventually call this principle into question.
As for the allocation of liability, the AI Act clearly divides it between the manufacturer, implementer, importer and distributor, which obliges companies to have compliance programmes that address the risks of AI, including the criminal liability of legal persons. Asimov’s Third Law translates here into post-market surveillance, corrective actions in the event of faults and up-to-date cybersecurity, as well as distribution contracts that include technical guarantees, a clear division of responsibilities, software updates and adequate insurance.
The author concludes by pointing out that Spain faces a regulatory jigsaw puzzle with pieces from different boxes – ranging from who pays if a robot causes harm whilst acting on its own, to whether specific technical standards or compulsory insurance are required; and that companies must analyse which regulations affect them, incorporate privacy by design and ensure they are fully prepared before entering a market whose rules are being written in real time.
Read the full article here