Skip to content

Legal Sandboxing and Policy Prototyping


  • Laura Lucaj (Volkswagen Group)

Currently, traditional legal frameworks seem to be struggling to keep up with the speed of development of AI systems. To address this challenge, policymakers and researchers have been exploring new ways of developing law. Practices such as legal sandboxing and policy prototyping have emerged as a way of enabling to develop laws that can effectively explore and address the real-world challenges emerging with the widespread deployment of AI.

The importance of regulatory sandboxes is recognised in the AI Act, as an opportunity to enable innovation advancement whilst understanding at an earlier stage the potential impact of systems once deployed, in order to develop proper mitigation measures in order to prevent such harms from occurring. However, the topic of regulatory sandboxes for AI systems is still under-investigated in the literature.

Regulatory sandboxes are not new, they originate in the field of computer science and have already been adopted in the attempt of understanding how to regulate financial technologies.

Regulatory sandboxes enable a form of experimental regulation that allows policymakers to understand how a product would behave in the market, to identify which risks it poses to society, in order to develop regulations that enable the effective prevention of potential harms whilst at the same time promoting innovations that benefit society.


In the AI Act, the limitation of traditional regulatory initiatives to face the challenges posed by AI is recognised. To address such concerns, the AI Act encourages the development of regulatory sandboxes to promote innovation and enable the creation of a supervised environment where experimentation and testing can be conducted. The scope of such environment would be to enable the development, testing and validation of AI system, hence, to combine the traditional elements of a sandbox with other methods such as piloting and testing.

However, currently the lack of detailed provisions for the implementation of sandboxes might result in the lack of incentive for the different Member States in establishing them

In spite of all the opportunities provided by sandboxes, such environments can be very costly to run. First of all, the expertise at the intersection of AI and law is being built and it is challenging and costly to set together such interdisciplinary teams.

Moreover, sandboxes by being usually run by public institutions, do often not have the necessary resources to address such complex challenges.

Policy Prototyping

What is policy prototyping?

Another emerging way of rulemaking is policy prototyping, which is the inversed process to legal sandboxing. Stakeholders come together in an environment where the policy is tested in the field in order to develop its final draft.

Such method originates in evidence-based policymaking and enabled the clear mapping of the scope a legislation should cover and all the test all the strengths and limitation of the policy proposal to avoid the costly impact on society of badly developed laws.

The process consists on the drafting of a prototype law, which is then tested by a group of stakeholders. Following the mock-compliance testing the involved participants can provide feedback to the policymakers on the shortcomings of the proposed regulation. Such feedback is essential in determining the effectiveness of the law and whether it is fit for purpose so that the necessary amendments can be taken.


Policy prototyping can significantly improve the field of AI compliance with law, by enabling to test whether the law can actually address its intended purpose in a real-world compliance simulation. Moreover, such method can account operational feedback that is hard to obtain through traditional ways, such as the public consultation initiatives undertaken by the European Union in the last years. In addition, policy prototyping can ensure that through testing and multiple iterations the scope of the regulation can be expanded and adjusted as needed. Lastly, it can increase trust and compliance with the regulation by offering a more transparent process of actionable legislation.


However, promising such method might sound, it entails several disadvantages that determine the low adoption on a larger scale. Conducting such experiments is very costly and can determine the insurgence of bias related to the inflexibility of changing the initial beliefs included in the first version. Moreover, the diversity of the group as well as their differing agendas can impact the results of the experiment and potentially decrease its effectiveness.

Last update: 2022.09.04, v0.1