A fresh instrument for assessing artificial intelligence systems has emerged from Luxembourg's national innovation hub. The Luxembourg AI Factory, which supports both commercial and public-sector organisations in developing, testing and rolling out AI solutions, unveiled the AI Assessment Sandbox Configurator (AASC) on 10 June. This open source platform tackles a longstanding bottleneck: the laborious, largely manual work required to verify whether AI models meet trustworthiness standards and comply with emerging regulations.
The Sandbox delivers a customisable testing environment for organisations committed to deploying AI responsibly. Daniele Pagani, who leads AI Research Strategy and Ecosystem Engagement at the Luxembourg Institute of Science and Technology (LIST), characterised the problem the tool addresses: traditional assessment involves "required low-value work before you can even start assessing the solution" because teams must manually locate and stitch together the necessary tests.
Trustworthiness as a Spectrum, Not a Checkbox
LIST, a research body specialising in natural and engineering sciences across Luxembourg and the Greater Region, partnered with the University of Luxembourg's Interdisciplinary Centre for Security, Reliability and Trust (SnT) to develop the tool. The platform allows any organisation to construct a bespoke AI testing framework by selecting from a curated, expandable catalogue of tests, controls and datasets tailored to their specific use case.
The Configurator includes a collaborative dashboard for reviewing and interpreting results alongside an automated report generator. Pagani framed trustworthiness not as a simple yes-or-no determination but as a multidimensional spectrum. The tool aligns with the European Commission's official trustworthiness framework, allowing users to browse the catalogue by dimension—transparency, explainability, fairness, robustness, safety, cybersecurity—by AI system type, by industry sector, or by regulatory authority.
For LIST, the Configurator represents the culmination of years spent operating its own AI assessment infrastructure. The decision to release it publicly stemmed from a desire to spare other organisations the effort of building such systems independently.
Bigger than Luxembourg, an Ambition for a Shared European Framework
The choice to publish the tool as open source reflects a deliberate strategy: transparency of the underlying code builds confidence among stakeholders, while distributing ownership across a broader ecosystem. Pagani articulated the vision as something akin to "the Hugging Face of AI testing." He elaborated: "It really has to become a community effort, European, maybe even beyond Europe. And the best way to make that happen is with open source."
Alessio Buscemi, Senior AI Scientist at LIST, outlined the multiple pathways for participation: "Contributions can take different forms. You can contribute to the code. You can contribute new control checklists or testing solutions. Or you can contribute by adopting the tool and providing feedback."
The Luxembourg AI Factory envisions the Sandbox evolving into a collaborative European resource, developed jointly across member states. Pilot programmes are already underway. Spain's National Competent Authority has contributed more than 15 checklists now incorporated into the system, while the OECD.AI repository, a worldwide platform advancing responsible and human-centred AI, has also contributed, extending the tool's reach across Europe and globally.
An Open Call for Organisations to Get Involved
The catalogue of testing and control mechanisms will expand to encompass additional sectors including healthcare, finance and manufacturing. Locally, the next phase involves launching a formal open call to engage public and private organisations of all sizes. The initiative will furnish technical support for deployment and usage while collecting input to refine and enhance the platform.
