AI Assessment Sandbox

The EU AI Act sets out requirements for AI systems based on their level of risk. Organisations will need to show that their systems are accurate, robust, secure, fair, transparent and subject to human oversight. Meeting these requirements is hard without the right tools: there are no universally agreed metrics yet, and many risks only appear when a system is tested systematically at scale. Biases, for example, can amplify discrimination far faster and more widely than traditional practices. 

The AI Assessment Sandbox, developed and operated by the Luxembourg Institute of Science and Technology (LIST) and SnT, gives you a controlled environment to evaluate and monitor your AI and multi-agent systems. It helps you find weaknesses early, reduce risks for users and build evidence of trustworthiness before you deploy.

It is based on LIST AI Sandbox, a hands-on testing environment where organisations evaluate AI models for robustness, fairness, bias, and regulatory compliance. That Sandbox has been put to work with organisations including Banque Internationale à Luxembourg, the City of Luxembourg, and Mistral AI.

The AI Assessment Sandbox is not a regulatory sandbox under the AI Act. It is a technical tool to help organisations prepare for compliance.

1. Run tests 

• Bias and fairness testing.
• Multilingual and Luxembourgish evaluation.
• Multi-agent behaviour assessment.
• Cybersecurity testing.

2. Monitor and report on results
All test results are collected in a single database with dashboards and reports. This gives you a consolidated view of your system's trustworthiness, and you can re-run tests to track improvements over time. Receive an independent assessment of your system by experts and scientists.

This service is intended for startups, SMEs, large companies and public administrations that are developing, integrating or deploying AI systems, including large language models and multi-agent systems, and want to assess their trustworthiness and prepare for the EU AI Act.