What are AI real-world labs, and what do they do?

With the AI Act, European lawmakers are attempting a difficult balancing act: On the one hand, effective regulation is needed for high-risk, emerging technologies that are already having significant societal impacts today. On the other hand, Europe should participate in the opportunities presented by these very technological developments—in other words, it should not simply slow them down, but rather shape them. AI real-world labs, or “AI sandboxes,” are intended to be one tool for this regulatory promotion of innovation: They are designed as special testing environments for AI products nearing market readiness, as a regulatory point of contact and source of support for companies and startups—and also as learning environments for the supervisory authorities, which must adapt their administrative processes to the nebulous concept of “AI.” The goal is to learn, both within and through the real-world lab, how to design AI applications in a way that meets the safety requirements of the AI Act, as well as how to intelligently incorporate the relevant testing criteria into administrative processes. This is no easy task for the EU member states, which do not have much time left to implement their first “AI sandboxes”: By August 2, 2026, these new types of regulatory bodies must at least formally exist and be operational.

Johannes Buchheim is a professor of public law and the law of digitalization at Philipps University of Marburg. In this Digitalgespräch, the expert on administrative law and legal frameworks in the digital society explains the role that AI real-world labs play in implementing the AI Act and what measures the EU is requiring its member states to take in this regard. He describes what forms of AI real-world labs might be conceivable for various specific technologies, who is involved in developing these public institutions, and what policymakers hope to achieve through them. Together with hosts Marlene Görger and Petra Gehring, Buchheim discusses what ideas, expectations, and fears might be at play among the stakeholders involved; how the framework conditions for AI real-world labs align with the competitive requirements of commercial enterprises; whether—and if so, under what conditions—participation in real-world labs is attractive to AI developers —and how transparency and disclosure requirements must be used to foster a critical public discourse in order to democratically legitimize this form of government-funded testing of potentially high-risk technologies.