Compliance

AI Regulatory Sandbox

Controlled framework for developing AI under regulatory supervision.

How the laws define this

All four instruments define an AI regulatory sandbox as a controlled, supervised environment where artificial intelligence systems can be developed, trained, tested, or validated under regulatory guidance to ensure safety and legal compliance. They consistently emphasize testing under authority supervision to mitigate risks and evaluate performance. However, these frameworks diverge in their primary focus and practical provisions. The AI Regulatory Sandbox Policy stresses mitigating risks to health, safety, and fundamental rights, whereas the Deutsches Nationales KI-Reallabor-Programm der Bundesnetzagentur highlights pre-market testing to examine legal questions under the AI Act. Meanwhile, the Estrategia Nacional de Inteligencia Artificial (ENIA) 2024-2027 emphasizes evaluating ethical impacts to inform future regulation by 2025. By contrast, the Lithuanian AI Act implementation amendments uniquely detail a statutory pilot run by the Innovation Agency with dedicated funding and technical support for market readiness.

Synthesised from the 4 statutory definitions below. Each one is quoted in full, with its source.

How jurisdictions differ

What is distinctive in each definition, binding law in force first.

Jurisdiction	Instrument	Status	What is distinctive
Lithuania	Amendments implementing the EU Artificial Intelligence Act: Amendments to the Law on Technology and Innovation (XV-105) and the Law on Information Society Services (XV-106) to implement Regulation (EU) 2024/1689 (Lithuanian AI Act implementation amendments)	In Force	Mandates a statutory sandbox operated by the Innovation Agency with timelines, funding, technical support, and capacity building for conformity and market readiness.
Germany	German National AI Regulatory Sandbox Program of the Bundesnetzagentur	Awaiting Entry	Focuses on development, training, testing, and validation even before market placement, based on an agreed sandbox plan to examine new legal questions.
Costa Rica	National Artificial Intelligence Strategy (ENIA) 2024-2027	Adopted	Proposes a time-limited testing environment by 2025 to trial solutions, evaluate safety, ethics, compliance, and inform future regulation and best practices.
Lithuania	AI Regulatory Sandbox Policy	Adopted	Specifies staged real-world tests under supervision by a competent authority to mitigate risks to health, safety, and fundamental rights.

What it triggers

In Lithuania, under Amendments to the Law on Technology and Innovation (XV-105) and the Law on Information Society Services (XV-106), AI system providers and deployers must use the AI regulatory sandbox to implement the EU AI Act.source
Lithuania's AI Regulatory Sandbox Policy sets out a supervised testing environment for AI developers and SMEs to guide compliance under the EU AI Act.source
Germany's German National AI Regulatory Sandbox Program of the Bundesnetzagentur describes a controlled environment based on a sandbox plan agreed with the competent authority.source

Definitions (4)

A supervised, time-limited testing environment, proposed for implementation by 2025, where innovative AI solutions can be trialed under controlled conditions to evaluate safety, ethical impacts, and compliance while informing future regulation and best practices.

Estrategia Nacional de Inteligencia Artificial (ENIA) 2024-2027Definition 1 of 4

A controlled environment supervised by a competent authority for the development, training, testing and validation of AI systems under an agreed sandbox plan, enabling staged real‑world tests and regulatory guidance while mitigating risks to health, safety and fundamental rights.

AI Regulatory Sandbox PolicyDefinition 2 of 4

A statutory, controlled pilot environment to be established and operated by the Innovation Agency to allow testing of AI systems under regulatory supervision, provide technical and compliance support to participants, and build capacity for conformity and market readiness; the amendments set timelines and funding for its establishment and operation.

Amendments implementing the EU Artificial Intelligence Act: Amendments to the Law on Technology and Innovation (XV-105) and the Law on Information Society Services (XV-106) to implement Regulation (EU) 2024/1689 (Lithuanian AI Act implementation amendments)Definition 3 of 4

An AI regulatory sandbox provides a controlled environment for the development, training, testing, and validation of innovative AI systems for a limited period, even before market placement. This occurs based on a sandbox plan agreed upon with the competent authority, aiming to examine new legal questions and ensure compliance with the AI Act and other applicable laws.

Deutsches Nationales KI-Reallabor-Programm der BundesnetzagenturDefinition 4 of 4

Not to be confused with

Digital sandbox

Digital sandbox refers more broadly to controlled experimentation with AI systems without specifying regulatory supervision or AI Act compliance.

Sandbox Plan

Sandbox plan is specifically the agreement document for sandbox participation, not the overall controlled framework.

AI sandbox / pilot

AI sandbox/pilot emphasizes time-limited experimentation, while this term includes ongoing supervised development and testing.

Questions and answers

What is required for AI regulatory sandboxes in Lithuania?

Amendments to the Law on Technology and Innovation (XV-105) and the Law on Information Society Services (XV-106) mandate an AI regulatory sandbox operated by the Innovation Agency. It provides technical and compliance support and builds capacity for EU AI Act conformity. The AI Regulatory Sandbox Policy guides supervised testing for developers and SMEs.

How does Germany's AI regulatory sandbox work?

The German National AI Regulatory Sandbox Program of the Bundesnetzagentur offers a controlled environment for AI development, training, testing, and validation before market placement. It operates based on a sandbox plan agreed with the competent authority to ensure AI Act compliance.

What does Costa Rica's AI strategy say about regulatory sandboxes?

The National Artificial Intelligence Strategy (ENIA) 2024-2027 proposes a supervised, time-limited testing environment by 2025. It allows trialing of AI solutions to evaluate safety, ethics, compliance, and inform future regulation.