Bringing AI legally compliant into practice: Reallabor KIRR Real develops exemplary AI and robotics solutions

Particularly, robotics in use without safety fences requires special safety measures. © Fraunhofer IPA / Photo: Rainer Bez

For example, the topic of explainability of AI using heatmaps can contribute to regulatory-compliant development. © Fraunhofer IPA/Photo: Rainer Bez

The goal of »KIRR Real: Living Laboratory for Legally Compliant AI and Robotics« of the Fraunhofer IPA and ARENA2036 was to support companies in Baden-Württemberg in implementing the EU AI Act as well as the EU Machinery Regulation coming into force in 2027. The results of the collaboration with 15 companies are now published and serve as guidance for companies with similar applications to develop AI and robotic systems in a legally compliant manner.

Time is of the essence when it comes to current regulations: On August 2, 2027, the obligations for high-risk AI systems as safety components will come into effect, and the entire EU AI Act will apply. The new EU Machinery Regulation also sets new requirements for safety-related AI and self-learning systems starting in January 2027.

Regulatory and technical consulting for companies

KIRR Real helped companies understand and implement these complex regulatory requirements. For this purpose, »Legal Quick Checks« were offered, which are short, concise consulting projects. A law firm classified these cases into one of the four risk classes of the AI Act (from minimal to unacceptable risk). Subsequently, the Fraunhofer Institute for Production Technology and Automation IPA examined the technical documentation and underlying concepts of the AI development to evaluate the legal compliance of the systems and to develop practical recommendations.

The project ran from November 15, 2024, to December 31, 2025, and was funded by the Ministry of Economics, Labour and Tourism of Baden-Württemberg. The consortium consisted of Fraunhofer IPA and the research campus ARENA2036, supported by several associated partners. The main objectives of the project were to investigate how well companies in Baden-Württemberg are prepared for the AI Act and the Machinery Regulation, how the two regulations interact and overlap, and where there are still hurdles in AI regulation.

The insights gained from the project provide impulses for practical implementation and further development of AI regulation and are actively fed back into the legislative process, for example through participation in standardization committees, regular exchanges with the Federal Network Agency, the Federal Institute for Occupational Safety and Health, and a workshop with the Federal Ministry of Labour and Social Affairs. The project results are also intended to influence the design of the living labs provided for in the AI Act.

Raising awareness of regulatory requirements

During the project, 15 AI application cases from areas such as manufacturing, humanoid robotics, healthcare, and personnel management were addressed. Even companies with advanced AI development still needed support in implementing the AI Act. Discussions also revealed that awareness of upcoming regulations is often insufficient. In applying the AI Act, either the legal or the technical aspect was frequently considered, but both aspects and early engagement with AI regulation are crucial for successful application.

Challenges in implementing the regulations

Particularly challenging for companies is the classification into risk categories. This classification is crucial for resource planning but can sometimes be ambiguous and depends heavily on the details of the application case. Therefore, guidance based on similar cases is not always possible.

The lack of harmonized standards means that more AI systems must undergo conformity assessment by third parties. Regarding the Machinery Regulation, many experts also see a need for clarification on what exactly constitutes a system with »self-developing behavior«.

Another issue is the technical implementation of the requirements. Many companies ask questions such as: How must evidence be provided? How extensive must documentation be? When is AI considered part of a machine? Handling multiple applicable regulations is often unclear and causes fear of unnecessary additional effort. Furthermore, many companies do not develop AI systems entirely in-house but use existing general-purpose AI models for their applications, which complicates the provision of the required evidence. However, the available guidelines and support tools have so far hardly addressed this practice.

An additional challenge for integrating AI into robots and other machines is posed by the new Machinery Regulation: companies must not only comply with the new rules for AI in safety functions over the next twelve months but also adapt their existing products to the new regulation.

Recommendations from the Living Laboratory

The experiences from KIRR Real show that while some companies are engaging with the regulations, existing uncertainties, the multitude of regulations to consider, and the lack of harmonized standards pose significant problems.

There is an urgent need for concrete, practical, and easily implementable guidelines, as well as structured approaches that start early in development. Particularly, AI development teams—regardless of company size—are often small and struggle to manage the required efforts. Support offers (such as living laboratories) should therefore be as accessible as possible to provide real added value. Companies should also receive support early in development, for example through legal assessments of an application case, workshops, or consulting during system development.

The Fraunhofer IPA will continue to collect insights from the project work and actively contribute to the legislative process, as well as use these insights for direct collaboration with companies. Companies can contact our representatives for this purpose. The goal remains to further develop AI regulation in a practical manner and to support companies in Baden-Württemberg optimally.