Today, answering that question is slow, expensive, and largely manual. Setting up a technical environment to assess AI means finding the right tests, standing up a sandbox, choosing the right metrics, and bringing it all together into results that technical, legal, and business teams can actually agree on. The EU AI Act adds real obligations on top of that — obligations most organisations don't yet have the tooling to meet efficiently.
Luxembourg just released an open-source answer.
The AI Assessment Sandbox Configurator
Launched in June 2026 by the Luxembourg AI Factory, the AI Assessment Sandbox Configurator is an open-source tool co-developed by the Luxembourg Institute of Science and Technology (LIST) and the University of Luxembourg's SnT in the framework of the Luxembourg AI Factory. It lets any organisation — a bank, a manufacturing SMEs. a public administration, a startup — build a customised environment for testing whether an AI system is trustworthy and compliant, in a fraction of the time and cost it would otherwise take.
At its core sits the catalogue of AI Tests & Controls, an open, extensible library of AI testing tools, controls, and datasets, built through an open plug-in API so the community can keep adding to it. It's already being pitched as a kind of “Hugging Face for AI Assessment” — a shared European asset rather than a proprietary product, with the potential to be endorsed by national competent authorities and to grow into sector-specific marketplaces of tests.
From that catalogue, an organisation selects the tests, controls, and datasets relevant to its own use case, and the Configurator generates a ready-to-use sandbox: an environment to execute the selected tests and controls, a collaborative dashboard to visualise the results, a report generator, and tamper-proof logs for auditability. Three things it's built to solve, directly:
- It adapts to any AI use case — every deployment has different risks, so the tool maps context and requirements to the right tests and metrics rather than applying a generic checklist.
- It brings results together and makes them comparable — all test results and control checks flow into one dashboard and report, aligned automatically.
- It aligns teams around a shared understanding — a common assessment structure means technical, legal, and business stakeholders are finally looking at the same picture.
Already at work, not just on paper
LIST's AI Sandbox work — of which the Configurator is the natural next step — has already been used with Banque Internationale à Luxembourg (BIL) to assess banking chatbots, with the City of Luxembourg on AI applications in public services, and with Mistral AI, one of Europe's leading AI developers. That's a regulated bank, a public administration, and a frontier AI lab, all already testing AI trustworthiness through Luxembourg-built, open-source tooling.
Open by design, sovereign by consequence
The choice to release this as open source isn't incidental — it's the point. Any organisation can install the Configurator in its own environment, extend the catalogue with its own tests and controls through the plug-in interface, or contribute to the tool's code itself. No vendor lock-in, no black box: an organisation can audit exactly what it's being tested against. That's what makes it a genuine contribution to digital sovereignty, not just a compliance shortcut — the ambition is a shared European infrastructure for AI assessment, built and owned collectively rather than licensed from a single vendor.
It's also part of a wider pattern. Luxembourg has spent the last several years building genuinely reusable open-source infrastructure out of public research — from BESSER's open-source low-code platform to CIRCL's threat-intelligence tools. The Configurator is the AI-trust chapter of that same story.
What's next, and how to get involved
The Configurator, its catalogue, and its source code are all available now on GitHub. Three ways in: use it to generate a sandbox for your own use case and tell the team what's missing; extend the catalogue with tests or controls relevant to your sector; or contribute directly to the codebase.
Curious what this could look like for your organisation?
- Discover the AI Assessment Sandbox Configurator
- Get in touch with the LIST dedicated team
- Come and meet us at the Open Source Conference Luxembourg 2026, where we'll be walking through the Configurator live.