Since August this year, the European AI Regulation has been in force. These laws and regulations were developed to manage the risks of AI and exploit the opportunities of AI without creating unnecessary obstacles. The AI regulation should not only prevent harmful systems from entering the market, but also aims to increase society's trust in AI through clear agreements on reliability and transparency.
The AI regulation is based on different levels of risk, with the rules of the regulation linked to the potential risks of AI systems or their applications. In this article, we address how innovative organizations can work with government and regulators to comply with these rules. Even and especially when it comes to entirely new AI systems or applications. Such as with so-called regulatory sandboxes.
Rules for AI
We ask Cas Mevissen, AI policy officer at the Ministry of Economic Affairs (EZ), for more information about a pilot for these regulatory sandboxes that is now underway from the Ministry of EZ together with Rijksinspectie Digitale Infrastructuur (RDI) and the Personal Data Authority (AP). But first some background: "The AI Regulation is European product regulation, which applies not only to stand-alone AI applications, but also when they are part of a product or security system. The law sets requirements for AI products on the European market to control risks to fundamental rights, human safety and health. The same approach has been used for this as, for example, for the safety of toys and medical devices, through the well-known CE mark," says Cas Mevissen.
Understanding regulations
A characteristic of high-risk AI systems is that they can have major impacts on people's lives, both positive and potentially negative. The AI regulation aims to allow for those positives, but reduce the risk of negative impacts. Some of the rules in the AI Regulation are complex and involve issues such as proper data management, risk management and transparency. And with the ongoing developments in AI, it is expected that there will also be organizations that do not fall readily into a category of the regulation.
"We expect that, in practice, companies will have questions about how some of the rules should be interpreted in the context of their product/service and what solution might best align with the rules. To this end, the AI regulation includes regulatory sandboxes as a tool. In doing so, regulators engage with organizations, especially small and medium-sized businesses, and provide support where possible so that they can comply with the AI Regulation with their innovative AI systems," Cas explained.
During the negotiations of the AI Regulation, the Netherlands pushed for the inclusion of a section on "regulatory learning. This means that the sandboxes must function as a feedback mechanism to develop best practices or FAQs, for example, or to test the AI regulation and adjust it where necessary in line with developments in the market. Participation in a regulatory sandboxes process provides a certificate of participation upon completion, summarizing what was discussed and done. This can then also be included in compliance reviews and any regulatory audits. According to the AI regulation, official regulatory sandboxes must be completed by August 2026.
Energetic launch of pilot
Work is underway in the Netherlands to pilot an AI question point that includes sandbox processes. The pilot phase kicked off at the annual conference of the Dutch AI Coalition in 2024, where AI developers were called upon to send in their questions about the AI regulation. Based on the dozens of submissions, the intended process approach is being tested. From the submissions, a selection was made of questions that can already be answered immediately (simple questions'), and submissions where more expertise, for example from sector-specific regulators, must be involved to formulate an answer; the so-called expert questions.
A small number of developers are also invited to go through a pilot regulatory sandboxes process. This is done for complex questions that require dissecting a case together with the AI developer. During this pathway, a series of in-depth discussions discuss the participant's often complex systems and how the AI regulation relates to them and how to comply with the regulation. No conclusions are drawn from this yet, but this is a way for regulators and questioners to practice with the regulatory sandbox. The insights from this pilot will then be used to properly set up regulatory sandboxes in the Netherlands.
The AP, RDI and the Ministry of Economic Affairs are working together in this pilot, also collaborating with other regulators. After all, supervision of the AI regulation requires the interplay of different supervisors, from different sectors and domains," said Cas. Among them are the Financial Market Authority and De Nederlandsche Bank, Healthcare and Youth Inspectorate and the Dutch Food and Consumer Product Safety Authority. Regulators other than those mentioned are also participating in the sandbox. The Ministry of EZ is also further collaborating with other ministries to set up the sandbox."
Learning process for all involved
The regulatory sandboxes are thus designed to help organizations understand how the rules of the AI Regulation apply to their situation and how to comply with them. This lowers the threshold for AI innovation in Europe, reduces the potential regulatory burden of having to comply with the AI Regulation, and strengthens the legal certainty of AI developers. But regulators and policymakers can also learn from the process.
"Regulators will soon experience in practice what does and does not work with the regulation and what new technologies organizations are working with. The general lessons that can be drawn from this can be shared with other market parties, regulators, but also with policymakers so that, where relevant, they can update the regulations in line with the new insights. This is how a learning ecosystem should emerge," Cas said. Similarly, the intention is to share, (always in consultation with the participating organization) conclusions and best practices from these trajectories with the AI ecosystem, so that other AI developers can benefit from them and know faster how to comply with the AI regulation. Recurring questions can be made into an FAQ.
Continued pilot
After the completion of the pilot phase, the Ministry of Economic Affairs and the regulators will evaluate the process. Then they will also discuss with relevant organizations, such as the NL AIC, how the findings fit with their experiences and knowledge of the AI ecosystem. Interested in more information? We will keep you informed of further developments through this website.
Roadshow AI regulation, are you joining too?
Are you working in the field of AI, does your organization use or develop AI systems? And are you curious what the AI regulation means for you or your organization? In cooperation with the Ministry of Economic Affairs, the NL AIC is organizing a roadshow of regional information meetings on the AI Regulation together with the regional AI hubs.
The Ministry of Economic Affairs will then take you through the main content and latest developments of the AI Regulation. Afterwards, a diverse panel of experts and entrepreneurs from the region will inspire you with examples and tips on what to expect in terms of regulations and the implementation of the AI Regulation. Keep an eye on our website and socials for the dates of these meetings.
In addition, are you interested in the activities of the Human-Based AI Working Group? Then visit this page.
