Beltug contributes to shaping Belgium’s AI regulatory sandbox

In April 2025, Beltug participated in this research through an online survey of some of our members and a meeting with the researchers, to help shape the future of the regulatory sandbox. We ensured that the perspective of real-world business users was included, raising important points around risk aversion, derisking compliance, and ensuring sandboxes are accessible to organisations of all sizes. We’ll keep you informed of any updates from the FPS Economy about the AI regulatory sandbox.

What is a regulatory sandbox for AI?

A regulatory sandbox is a controlled environment that allows for the safe development, testing, and validation of AI systems before they are deployed in the real world. The purpose is to encourage innovation, provide regulatory clarity, and enable responsible AI adoption, even for complex or high-risk use cases. The sandboxes operate under the supervision of regulators, but with temporarily relaxed or adapted rules.

Key takeaways from Beltug’s contribution

12 Beltug members participated by filling in the online questionnaire from VUB and FARI, and 4 took part in the in-depth discussion with the researchers at the end of April 2025. Some of the takeaways of their participation included:

1. AI adoption is widespread, but organisations are hesitating on high-risk use 
   All of the members who completed the questionnaire are using or planning to use AI, especially in areas like generative AI, natural language processing, computer vision and predictive analytics. However, they are more reluctant to explore high-risk AI applications, such as managing traffic systems or evaluating applicants.
2. The leap from PoC to deployment is a challenge 
   The members voiced a need for support in moving from proof-of-concept (PoC) stages to large-scale deployment. The primary hurdle is the uncertainty and risk around compliance and liability. Clear assurance and assistance in the process will help get AI rolled-out and applied more quickly and on a larger scale.
3. Derisking compliance is essential 
   While some participants indicated they have resources for handling regulatory compliance, they are not fully confident that they understand their obligations under the AI Act.
4. Perceived benefits of sandboxes 
   The Beltug members agreed on the different potential benefits of participating in sandboxes, such as better understanding the regulatory environment; faster, less expensive compliance; enhanced customer trust; and earlier feedback supporting safer AI solutions. This makes it clear that all these benefits could be delivered, depending on how the sandbox is set up.
5. Beyond SMEs
   We strongly argued that sandboxes must be accessible to large organisations as well as SMEs. The AI Act provides for free access to SMEs, but larger players should also be able to participate, even if this involves some financial contribution.
6. Hosting concerns
   Most AI sandboxes are hosted in cloud environments and tailored for regulatory or compliance testing. However, these are run on US cloud environments, which raises data sovereignty, privacy and security considerations. As a result, it is more likely that hosting will be done on-premise using a national or institutional data centre. Another possibility is a hybrid approach with local data storage but compute offload to the cloud.
7. Archiving and version control of AI models
   Participants raised important questions about the evolution of AI models within the sandbox. What happens when a model is updated or repurposed? Should only the new components be re-tested, or the entire system?
8. IP concerns and expert involvement
   Concerns about intellectual property (IP) ranked high, yet organisations were open to allowing expert involvement from academia or industry, provided that clear contractual agreements are in place to protect proprietary technologies.
9. Stakeholder involvement
   Beltug emphasised the importance of the continuous involvement of future sandbox users. We hope this message is clearly reflected in the final report submitted by VUB and FARI to the FPS Economy.

What’s next?

It has not yet been confirmed which authority will manage the regulatory sandbox in Belgium. As the current telecom and post regulator, the BIPT would be in charge, but no final decision has been made.

Despite the many questions still on the table, Beltug members take a positive and constructive view of the regulatory sandbox initiative. If designed well — with regulatory clarity, swift decision taking and good stakeholder consultation — it could become a key enabler of responsible AI innovation in Belgium. As always, we’ll keep you informed.