Implementing Regulation (EU) 2026/2487 lays down the rules applicable to artificial intelligence regulatory sandboxes in the European Union, covering access, application requirements, confidentiality, data protection, risk management and the recognition of testing results. Small and medium-sized enterprises (SMEs), including startups, benefit from priority and free access to certain regulatory services, while providers and prospective providers may apply to sandboxes in other Member States. The Regulation also provides for sandbox plans, exit reports, extension and suspension mechanisms, and the establishment of national sandboxes by 2 August 2027, without relieving companies of their responsibility to comply with applicable legislation.
November saw an important development under the European Union’s Digital Omnibus Package relating to artificial intelligence (AI): the adoption of Implementing Regulation (EU) 2026/2487 of 10 November 2026 (Regulation), which lays down rules on the establishment, development, implementation, operation and supervision of AI regulatory sandboxes.
Key aspects include access arrangements, especially for small and medium-sized enterprises (SMEs), and the safeguards regarding confidentiality and risk management, as well as the recognition of the results obtained, although this does not, however, amount to a declaration of conformity. The key measures are set out below:
- Priority and free access for SMEs – SMEs, including startups, are given priority and free access to regulatory guidance, advice and supervision services. Other services may incur exceptional costs, which must be communicated in advance and recovered in a fair, transparent and proportionate manner (see Article 2(3) of the Regulation).
- Applicants from third countries – Applicants without a registered office or branch in the European Union must pay the applicable fee and appoint, in writing, an authorised representative in the Union, with whom they must jointly submit their application (see Articles 2(5) and 3(2)(a) of the Regulation).
- Access and application requirements – Providers and prospective providers may apply to sandboxes in other Member States and are not limited to those in the country where they are established. Systems not yet placed on the market or put into service are eligible, as are, under specific conditions, existing systems undergoing a substantial modification or a modification that renders them high-risk. The selection process takes into account, among other factors, the regulatory challenges, the project’s maturity and innovative nature. Applicants must be notified of the decision, which may be appealed, within three months at the latest following the submission of a complete application or the closing date of the call for applications, as the case may be (see Article 3(3) and (7) of the Regulation).
- Confidentiality and trade secrets – All parties involved are subject to confidentiality obligations. The sharing of technical tools and results must respect intellectual property rights and protect confidential information and trade secrets (see Article 4(3) and (6) of the Regulation).
- Sandbox plan – Following selection, the participant and the authorities agree on a plan identifying the applicant and the AI system, setting out the timetable, objectives, activities, responsibilities, services and costs. The plan must cover the legal requirements to be assessed, the processing of personal data and whether the data protection authority needs to be involved and, where appropriate, the management of risks and serious incidents, as well as real-world testing. The participant must identify, as far as possible, any confidential information (see Article 5(1) and (2) of the Regulation).
- Exit report – Issued within 45 working days after the end of participation, the report and written proof of the activities are recognised by market surveillance authorities and notified bodies throughout the Union, but do not constitute a declaration of conformity. Publication of the report is subject to the express consent of the participant and the authority; non-confidential conclusions may be disclosed without such consent. Information not marked as confidential in the plan or before the end of the project is presumed to be available for disclosure, without prejudice to the rules on access to documents (see Article 6 of the Regulation).
- Extension and suspension – The duration of the project may be extended by justified agreement with the authority, with the decision to be taken at least 30 calendar days in advance. Testing or participation may be suspended temporarily or permanently if it is not possible to implement mitigation measures, particularly where there are significant risks to health, safety or fundamental rights (see Article 8(1) and (2) of the Regulation).
The following table sets out the main deadlines and milestones:
|
Date |
Description of the milestone |
|
1 December 2026 |
Entry into force of the Implementing Regulation on the twentieth day following its publication on 11 November 2026 (see Article 10 of the Regulation). |
|
2 August 2027 |
Deadline for Member States to ensure that national sandboxes are in place (Recital 6 of the Regulation). |
What remains unchanged
Participation in a sandbox does not relieve the company of its responsibility to comply with applicable legislation, including the rules on real-world testing.
Next steps
Economic operators should, in particular:
- Assess and prepare for participation in AI sandboxes, including those in other Member States, taking into account the eligibility and maturity of projects and, in the case of SMEs, the conditions for priority and free access. It is important to define the testing objectives, the regulatory issues to be clarified, the resources, costs and responsibilities, whilst anticipating the needs for personal data protection, risk management and the identification of confidential information.
- Integrate experimentation into the compliance strategy, using the test results and exit documentation to support the assessment of compliance with applicable requirements.
The Technology team at Morais Leitão will continue to monitor legislative and regulatory developments in the artificial intelligence sector closely and remains available to answer any further questions.