Glossary · AI Act
Regulatory Sandbox
A controlled environment, run by authorities, where AI systems can be developed and tested under regulatory supervision before market launch.
Explained in depth: AI Act overview
The AI Act requires every member state to establish at least one AI regulatory sandbox (alone, jointly with other member states, or via participation in an existing one) by 2 August 2026. In a sandbox, providers develop, train, and test innovative AI systems for a limited period under the guidance of the competent authority, with a sandbox plan agreed up front. Authorities provide guidance on regulatory expectations, and documented sandbox participation can support later conformity assessment. The act ties incentives to sandboxes, particularly for SMEs and startups: priority access, reduced fees, and support in producing the required documentation. Under specified conditions, sandboxes also allow testing under real-world conditions.
Why it matters
For companies building borderline or novel high-risk AI, a sandbox converts regulatory uncertainty into a documented dialogue with the authority, valuable both for engineering decisions and as evidence of good faith if questions arise later.