What happened
The European Commission adopted Implementing Regulation (EU) 2026/2487 on November 10, 2026, published it November 11 and set entry into force for December 1. It adopts common operating requirements for supervised AI regulatory sandboxes: a place for providers to develop and test an eligible system with regulators under an agreed plan before sale or qualifying modification. The regulation specifies eligible project categories and selection criteria, covered supervisory services and exceptional-cost disclosure, admission-decision appeals and written plans allocating activities, responsibilities and risk safeguards. Public bodies receive the specified favorable pricing treatment. An exit report must be provided no later than 45 working days after completion; it is not market approval or certification.
The original AI Act already required transparent selection, three-month admission decisions and free small-firm access subject to exceptional costs. Those are not newly created benefits. Admission remains subject to eligibility and regulator capacity; adoption does not establish a fully staffed operating sandbox or access for every applicant.
Potential direction of power
Leaning dispersion · Dispersion +1
Binding eligibility, selection, fee and supervisory requirements make the regulatory route more predictable and reviewable for smaller providers and public institutions. They constrain administrative discretion and can reduce reliance on expensive private compliance expertise. Actual access still depends on eligibility and public capacity.
Window of contestability
Opens
Binding eligibility, selection, fee and supervisory requirements make the regulatory route more predictable and reviewable for smaller providers and public institutions. They constrain administrative discretion and can reduce reliance on expensive private compliance expertise. Actual access still depends on eligibility and public capacity.
