The European Commission launches consultation on draft rules for AI regulatory sandboxes under the EU AI Act

The European Commission has begun consulting on the draft Implementing Regulation on Artificial Intelligence Regulatory Sandboxes, inviting feedback until 30 December 2025. The proposal sets out the operational framework for the creation, management, and supervision of AI regulatory sandboxes established under the EU AI Act. These sandboxes are intended to provide controlled environments where organisations can develop and test innovative AI systems under regulatory oversight.

The draft text defines the terms and conditions for participation, with a clear emphasis on supporting Small and Medium-sized Enterprises, including start-ups. SMEs would be able to join sandboxes free of charge, while larger companies may be subject to cost-recovery fees. The proposal also outlines the application and selection process, limiting eligibility to AI systems that have not yet been placed on the market or that are undergoing significant modifications.

Competent authorities would be required to document sandbox activities, publish relevant compliance tools, and agree on a detailed sandbox plan with each participant. This plan must set out objectives, timelines, methodologies, and safeguards to mitigate risks during testing. At the end of participation, authorities must issue a written proof and prepare an exit report to capture regulatory insights. These documents serve only as evidence of participation and learning; they do not constitute a declaration of conformity under the AI Act.

The draft regulation also includes provisions for suspending projects that present safety or fundamental rights concerns, and it obliges authorities to publish annual reports on sandbox activities to strengthen transparency and regulatory learning across the EU.