AI regulations do not ban or allow robot force

European rules on artificial intelligence and machines do not clearly ban or authorize autonomous robots from using force. Analyses of the AI regulation (RIA) and the "machines" regulation show that a robot's AI system is not automatically considered "high risk" under these rules. The RIA defines high-risk AI systems based on specific conditions, such as being a safety component of a product covered by Annex I and being subject to third-party conformity assessment.

High-risk AI criteria are specific

The RIA narrows the concept of safety components but retains those whose failure could endanger health and safety. A system that decides to strike a human would fall into this category, but whether such a system would be subject to third-party assessment depends on its design. The regulation does not explicitly address the use of force by autonomous systems.

Machines regulation covers certain safety functions

The "machines" regulation requires the involvement of a notified body for six categories of products, including systems with self-evolving behavior. However, it is unclear whether neutralizing an aggressor would be considered a "safety function" under this regulation. The Commission's draft guidelines take a broad view, but the final version is not expected until the end of 2026.

RIA requirements do not apply directly to machines

Even if a robot's AI system is classified as high risk, it would still escape the essential part of the RIA. The "machines" regulation is listed in section B of Annex I of the RIA, meaning only specific articles apply to it. None of the requirements of Chapter III of the RIA, including the human control requirement, are directly applicable.

Annex III covers perception, not action

Annex III of the RIA does not specifically target the decision to use force, but it can cover the perception chain of a security robot, such as remote biometric identification and emotion recognition. These perception modules would be subject to certain articles of the RIA starting December 2, 2027. However, the decision to strike would remain out of scope.

Machines regulation excludes weapons

The "machines" regulation already addresses autonomous mobility but not the use of force. It requires specific risk assessments, control systems, and supervision functions for autonomous mobile machines. However, it was not designed for machines intended to exert force on people. If a robot were considered a weapon, it would fall outside the scope of the "machines" regulation, which excludes weapons and machines designed for military or police purposes.

Implementation timeline and legal limits

The schedule for the implementation of these regulations includes the entry into force of the "machines" regulation on January 20, 2027, and the application of Annex III of the RIA on December 2, 2027. The deadline for delegated acts under the RIA is August 2, 2028. According to these texts, nothing prohibits the placing on the market of an autonomous humanoid robot compliant with the "machines" regulation, but nothing authorizes it to use force either. These are market placement legislations: they state under what conditions a product can be sold, not what it can do once on the street.