EU adopts rules for AI Act regulatory sandboxes
Implementing Regulation 2026/2487 sets selection, sandbox-plan and exit-report rules, with priority and free access for small businesses; it was published in the Official Journal on 11 November.
- Government & policy
- Minor
The European Commission adopted Implementing Regulation (EU) 2026/2487 on 10 November, setting common rules for how AI regulatory sandboxes under Article 58 of the AI Act are established, run and supervised. It was published in the Official Journal on 11 November and enters into force on the twentieth day after publication.
Sandboxes are supervised environments in which providers can develop, train and test AI systems with a regulator before placing them on the market. The regulation says common rules are needed to avoid fragmentation across member states and to allow results to be mutually recognised.
Among its provisions, selection procedures must be transparent and fair, and applicants must be told the outcome within three months of a complete application. Small and medium-sized enterprises, including start-ups, get priority access and free regulatory guidance, while other participants may be charged the full cost of services. Each participant agrees a sandbox plan with the supervising authority, and on completion receives written proof and an exit report. The exit report is explicitly not a declaration of conformity, and the recitals say participation does not amount to approval to place a system on the market. Market surveillance authorities in any member state must recognise the report as documentation of what was done.
It is a technical measure that follows the Digital Omnibus changes to the AI Act and the first harmonised standard.