New Sandbox Report Published

Why a Sandbox?

Due to the rapid pace of technological progress, the regulatory environment for AI technologies is often unclear for businesses, research institutions, and public administrations. As a result, uncertainties arise when implementing AI projects, which can hinder innovation. The Innovation Sandbox provides a testing environment where stakeholders can carry out AI initiatives within a clearly defined framework.

Start-ups, SMEs, large enterprises, and research institutes gain access to regulatory expertise and novel data sources through the sandbox, enabling them to foster innovation and support data-driven developments. Collaborative work within the sandbox also ensures effective knowledge transfer between participating organisations. All insights and results are shared publicly and serve as valuable input for shaping future legal frameworks.

Unlike many international approaches, the Innovation Sandbox for AI goes one step further: submitted projects are not only reviewed but also implemented in practice.

The goal is to promote responsible innovation that takes legal and ethical considerations into account, and to sustainably strengthen the use of AI in public administration, business, and research.

Goals

Advantages:

• Provide regulatory clarity

Advantages:

Advantages:

Advantages:

• Foster innovation and offer access to data

Advantages:

Advantages:

Advantages:

• Ensure knowledge transfer

Advantages:

Advantages:

Advantages:

• Deliver input for future legal frameworks
  

Phase I: 2022-2024

The first phase of the Innovation Sandbox for AI was successfully completed in March 2024. The insights gained from five AI projects have been published.

Phase II: Current Projects

Between mid-April and the end of May 2024, a project call was held, during which 24 organisations submitted AI project proposals. The Sandbox Steering Committee selected the projects based on the criteria listed below. The project team is currently working on six new sandbox projects, and the results will be published on an ongoing basis.

AI in Medical Documentation

In healthcare, documentation requirements generate a lot of administrative work. This project shows how speech recognition and large language models (LLMs) can reduce this workload. At the same time, it examines the legal framework – for example, with regard to data protection, professional secrecy and cloud use – and clarifies when such systems are considered medical devices. The aim is to develop concrete best practices for the safe and responsible use of AI in medical documentation.

EU Regulation of Autonomous Inspection Robots

For many Swiss robotics companies, access to the EU market is crucial – yet regulatory complexity is increasing. This project analyses key legislative frameworks such as the AI Act, the Machinery Regulation, the Cyber Resilience Act, and the Data Act, and demonstrates how companies can effectively navigate these various requirements. In addition, the project tests AI governance software designed to help firms efficiently implement technical, legal, and organisational standards – for example, those set out in ISO 42001.

AI Diagnostics in Ophthalmology

Diagnosing eye diseases is complex and requires specialist expertise. This project develops an AI-based approach for detecting diabetic retinopathy and investigates how AI can be reliably integrated into clinical routines. In addition to medical advancements, the project offers practical recommendations that can be transferred to other disciplines such as radiology or dermatology.

AI-Based Review of Building Applications

Building permit procedures are often lengthy and complex. In collaboration with a city administration, this project develops a prototype that uses multimodal AI models to automatically review building applications – analysing both textual documents and construction plans. This allows regulatory compliance to be assessed more efficiently. The results provide valuable input for municipalities and solution providers in driving digital transformation in the construction sector.

Combating Deepfakes in Identity Fraud

Deepfakes make it increasingly easy to forge identification documents and create fake identities. This project explores how organisations can securely and effectively share knowledge, signals, and data about fraud attempts. The aim is to detect fraudulent patterns early and prevent identity fraud involving generative AI – even across organisational boundaries.

Bridge Maintenance Using AI Sensors

Many bridges are decades old and still primarily inspected visually. This project tests AI-powered sensors that continuously monitor the vibration behaviour of bridges. This enables early detection of structural fatigue and allows for informed assessments of load-bearing capacity. The data-driven approach aims to extend the service life of bridges and make maintenance more efficient.

Sandbox Initiators

A wide and diverse group of institutions and individuals from public administration, research and the private sector have come together to launch the sandbox initiative. The aim is to strengthen Zurich as an innovation location for artificial intelligence. This initiative is receiving significant financial support from the Zurich Metropolitan Area.

The following institutions are part of the initiative: 

Publications

In the Media