The EU sets common rules for AI Act regulatory sandboxes
On 10 November 2026 the European Commission adopted Implementing Regulation (EU) 2026/2487, which sets out how the regulatory testing environments — the sandboxes — provided for in the Artificial Intelligence Act, Regulation (EU) 2024/1689, must be set up, developed, operated and supervised. According to the official timetable, these implementing rules apply from 1 December 2026. Member States must have at least one national sandbox up and running by 2 August 2027, and may set it up jointly with other countries.
What an AI sandbox is — and what it is not
A sandbox is not an “approved product” seal. It is a controlled environment, with a plan agreed between the provider (or prospective provider) and the competent authority, for developing, training, testing and validating AI systems for a limited time before placing them on the market or putting them into service. The implementing regulation aims to ensure these tests follow common criteria across the Union, so that results are interpreted consistently from one Member State to another.
There is a nuance worth noting. The written proof the authority may issue at the end of participation documents what was done in the sandbox, but it does not carry the same legal value as a declaration of conformity under Article 47 of the AI Act. Taking part helps prepare for compliance and reduce uncertainty; it does not replace the provider’s obligations or remove its liability if the system causes harm.
What this means for a Spanish SME
Most SMEs do not train foundation models. They are, however, increasingly integrating AI systems into customer service, invoicing, human resources, logistics or e-commerce. If the system may be considered high-risk — or if there is reasonable doubt about its classification — a national sandbox can be a way to test it under supervision, clarify obligations and document safeguards. It is also useful when a substantial change is about to be made to a system already in use.
August 2027 may seem far away, but it is not if you need to inventory AI uses, assess risks, draft policies and prepare test data. In Spain it is worth keeping a close eye on which competent authority will run the sandbox and when applications will open. In the meantime, compliance work should not wait: the AI Act is already applying in phases, and documentation — inventory, governance, transparency, human oversight — is required whether or not a testing environment exists.
At Digitacode we help companies integrate AI into their processes with technical and legal judgement. If you need an inventory of systems, a risk assessment or a compliance plan, LeydeIA is designed precisely for that work.
Sources
- Implementing Regulation (EU) 2026/2487 of 10 November 2026 (EUR-Lex).
- EU Publications Office record.
- EFTA record 32026R2487 (date of adoption and of application in the EU).
- Article 57 of the AI Act, European Commission AI Act Service Desk.