The City of Luxembourg plans to deploy a chatbot for its citizens. This citizen-facing chat agent must be multilingual and able to answer questions about the widerange of municipal services. Before launching a call for tender, the City conducted three POCs, to assess the feasibility of introducing such a chat agent and partnered with the Luxembourg Institute of Science and Technology (LIST) to make citizen-facing AI safer and fairer. City of Luxembourg brought one of these multilingual chat-agent platforms into the LIST AI Assessment Sandbox to validate feasibilty and risks before moving ahead.
The solution (assessment-first)
The City of Luxembourg explored citizen-facing, multilingual chat AI agents that could answer questions across a wide range of city services. The concept entered the AI Assessment Sandbox to measure fairness before any large-scale build or procurement.
Inside the assessment: how we tested
- Bias & fairness test suite: We ran hundreds of prompts across three languages to evaluate demographic parity and content consistency for the selected innovator’s solution.
- Scenario coverage: Prompts mirrored realistic resident interactions about city services, including sensitive contexts.
- Multilingual parity checks: We checked consistency of responses across languages to detect any unequal treatment.
- Evidence pack: Results consolidated into a reproducible report to support decision-making.
What we found
The first assessment revealed discrimination patterns related to gender and nationality. Based on these findings, the chatbot provider activated the necessary guardrails in their system to eliminate the observed biases. We then re-tested to confirm improved fairness while keeping responses helpful and on-topic. The City’s feedback was positive. Our assessment helped them understand the risks associated with chat agents and make an informed decision.
How LIST’s SmartCityHub supports cities end-to-end (Citcom TEF site)
SmartCityHub is LIST’s Testing & Experimentation Facility (TEF) for smart cities under Citco. It offers a portfolio of services—and the AI sandboxes are one part of that offer. The services are organised into three stages that cities and solution providers can move through.
- Guided Exploration
Scope city needs and opportunities; match providers’ solutions to real missions and identify pilot partners. - Technical Experimentation
Run structured experiments in a safe environment at the TEF site—cities validate concepts with real data; providers test products in real contexts and gather credible evidence. - Trusted Adoption
Move from pilot to procurement with neutral evaluations—interoperability, data protection, and AI Act checks—plus clear documentation to support compliance-ready, trusted deployments.
Beyond the sandbox: a common language for trust
To harmonize how experiments are evaluated across cities, LIST launched an AI Assessment Club in April 2025 (Citcom T3.2). Through facilitated design-thinking sessions (six completed to date), the Club is defining a Citcom label and result-specific, tamper-proof badges linked to the Citcom Hub—backed by harmonised guidelines and non-binding compliance recommendations to guide cities and procurement officers. Evaluations culminate in harmonised reports with legal disclaimers and public visibility for participating innovators.
Why engage with LIST’s Smart City AI Sandbox
- Assessment before investment: Independent testing to validate fairness, safety, and usefulness—then decide what to scale.
- EU-ready by design: Evidence artifacts and reports that help with alignment with emerging EU AI Multilingual equity: Purpose-built checks for European linguistic reality to reduce response gaps.
- Vendor-neutral & open: Interoperable methods, transparent reporting—no lock-in, clearer procurement.
- Reusable playbook: A phased path (guided exploration => Technical experimentation => Trusted adoption) you can apply across use cases.
Why it’s reliable
At the TEF Site Luxembourg, cities don't just test technology—they build trust. Every experiment is grounded in transparency, security, and compliance, ensuring cities can confidently innovate while maintaining complete control. Here’s why:
- Data ownership: Partner cities retain full ownership of their data and solutions.
- Regulatory compliance: As a TEF, we help with adherence to national and EU data and AI regulations.
- Transparency: A structured evaluation process with full documentation at every stage of the experiment.
Luxembourg City shows how pragmatic, assessment-first governance translates into better AI services for residents. With the right sandbox, safeguards, and shared standards, cities can move from pilots to confidence—without slowing innovation.
Want to explore your own AI assessment journey?
Get in touch with LIST to start a scoped experiment for your city’s AI services: smartcityhub@list.lu
