Skip to content
← All insights
EU AI Act
4 min read

Regulatory Sandboxes and The EU AI Act

The EU Artificial Intelligence (AI) Act aims to lead the world in the governance of AI, requiring impact assessments to identify the risk associated with the use of AI systems and continuous management and mitigation of this risk.

Regulatory Sandboxes and The EU AI Act

Updated November 2026: The EU AI Act is in force, and the Digital Omnibus on AI requires every member state to have an AI regulatory sandbox by 2 August 2027. The sections on the 2022 Spanish pilot are kept for historical context.

The EU Artificial Intelligence (AI) Act aims to lead the world in the governance of AI, requiring impact assessments to identify the risk associated with the use AI systems, and continuous management and mitigation of this risk. One of the major contributions of the EU AI Act is the creation of regulatory sandboxes, which provide participants of the initiative the opportunity to undertake controlled experiments and testing of their products under the supervision of relevant authorities. The sandboxes therefore provide an environment for providers to test the compliance of their product before it is launched on the market.

What are EU AI Act regulatory sandboxes?

The EU AI Act gives small and medium-sized enterprises (SMEs), including start-ups, priority and free-of-charge access to the sandboxes (Article 62), removing some of the barriers they may face when launching a product. However, participants remain liable under applicable Union and national law for any harm inflicted on third parties as a result of experimentation in the sandbox. That said, where they observe the sandbox plan and follow the authority’s guidance in good faith, no administrative fines are imposed for infringements of the AI Act (Article 57(12)).

The Act requires each member state to establish at least one AI regulatory sandbox at national level, by 2 August 2027 following the Digital Omnibus (originally 2 August 2026). The AI Office may also establish a sandbox at Union level, to facilitate the EU’s vision, and calls for common rules to be established to promote more standardised approaches across the member states, and to facilitate cooperation between overseeing authorities. Ahead of the Act, Spain announced that it would pilot a regulatory sandbox aimed at testing the requirements of the legislation, as well as how conformity assessments and post-market activities may be overseen.

Deliverables from this pilot include documentation of obligations and how they can be implemented, and methods for controlling and following-up that can be applied by those supervising national authorities responsible for implementing the regulation.

The Spanish pilot (2022)

Reflecting the cross-state approach desired by the Act, other member states will be able to follow or join the pilot. The pilot was planned to begin in October 2022, with results expected by the end of 2023. With a budget of 4.3 million euros, the pilot will be financed by the Spanish government’s recovery and resilience funds as part of the Spanish National AI strategy.

Sandboxes are expected to have benefits to both businesses or providers, who are able to develop and test their products in a realistic environment to ensure that they meet the appropriate regulations, and to regulators, who are better able to understand the products they are regulating. This can result in a shorter time to market, with the products that are released potentially being safer for consumers.

However, sandboxes can also have limitations, including the risk of abuse. For example, those with harmful intentions could misuse the sandboxes, meaning that the regulation is made less stringent as a result, potentially resulting in harmful products being released on the market. Further, the use of sandboxes could delay innovation from private actors if the actions of regulators harm their research and development processes.

Notwithstanding these concerns, the final Act made sandboxes mandatory: every member state must now operate at least one, and the Commission adopts implementing acts setting common rules for their operation (Article 58). Providers can also test high-risk systems in real-world conditions outside a sandbox under Article 60, which the Digital Omnibus extended to Annex I products.

Why a regulatory sandbox for AI is needed

Proponents believe that the AI sandbox provisions of the Artificial Intelligence Act, which aim to enable controlled experimentation between AI developers and regulators, will be beneficial for AI regulation in Europe because it will facilitate responsible AI development in EU member states. By providing controlled environments for companies to test compliance with emerging legal frameworks, it is argued that the Act gives developers the chance to refine AI systems under real-world conditions, collaborating with authorities to demonstrate compliance. Some commentators, however, have suggested that there should be more interplay between AI sandboxes and the regulatory frameworks outlined by the data protection authorities.

Holistic AI – clearing a pathway towards EU AI Act compliance

As AI regulation accelerates across the EU and globally, the focus on fairness and harm reduction is intensifying. These are more than buzzwords at Holistic AI – they're the core of what we do. By merging technical expertise with a deep understanding of policy, we ensure a comprehensive grasp of both the technology and its application in various settings.

Holistic AI offers EU AI Act compliance solutions, including an automated readiness assessment and an agentic AI governance platform that runs framework assessments against the EU AI Act.

Before applying to a sandbox, use Holistic AI’s EU AI Act readiness assessment to classify your AI systems by risk, map your obligations and evidence your compliance level, which is the kind of documentation sandbox applications ask for.

Last updated November 2026.

KEEP READING

More insights from Holistic AI.

EXPLORE IDENTIFY

Give your AI a starting point.

Discover how systems, ownership and evidence connect.

Explore Identify
↑