An AI system used by a municipal welfare agency to score citizens' eligibility for housing subsidies based on…
In accordance with our privacy statement.
The municipal welfare agency is a public authority deploying an in-house AI system that explicitly includes a social-scoring component evaluating "lifestyle patterns" to determine housing subsidy eligibility. This directly matches the prohibited pattern under Art. 5(1)(c): a public authority using AI to score natural persons in ways that produce detrimental treatment (denial or reduction of housing benefits) where the lifestyle-pattern data is arguably unrelated to the original purpose of financial need assessment, and where weighting such patterns may be disproportionate to the underlying behaviour being measured.
- 1Immediately isolate the social-scoring / lifestyle-pattern component from the rest of the scoring model and have legal and technical teams document precisely what data inputs feed it, what weight it carries in the final score, and whether those inputs were collected for this purpose.
- 2Commission a legal opinion specifically on whether the lifestyle-pattern sub-component meets the disproportionality and unrelated-data tests in Art. 5(1)(c), treating the output as privileged and not a green-light for launch.
- 3Map every input variable in the full scoring model against its original collection purpose to identify which variables could be characterized as social-scoring indicators rather than financial-need proxies.
- 4Audit the caseworker override rate and document it formally. A near-zero override rate means the AI output is the operative decision, which increases the agency's exposure as the functional decision-maker under the prohibition.
- 5If the social-scoring component cannot be cleanly excised, suspend use of the full system for benefits decisions until a redesigned model without that component has been reviewed.
- 6Prepare a stakeholder brief for the agency director explaining that deployment in current form carries a plausible prohibition risk under Art. 5(1)(c), not merely a high-risk classification issue, so that the decision to proceed or pause is made at the appropriate level of authority.
Interpretive points where reasonable counsel would disagree, applied to the specifics above.
- The system evaluates 'lifestyle patterns' as part of housing subsidy scoring. Whether those pattern inputs were collected originally for welfare-need assessment or were derived from other administrative datasets (e.g., utilities, mobility, social services records) is dispositive for the unrelated-data prong of Art. 5(1)(c), and that factual question is unresolved.
- The scoring is described as the 'primary basis' for decisions with caseworkers rarely overriding. At what point does a nominally human-in-the-loop structure become, in practice, automated decision-making by the AI system such that the agency cannot argue the AI is merely a tool rather than the operative scorer?
- The February 2025 Commission Guidelines draw a narrower line around lawful credit scoring and insurance pricing. The team will need to determine whether the financial-history and employment-pattern sub-components of this model are sufficiently analogous to lawful credit modelling to be carved out, or whether their combination with lifestyle-pattern scoring taints the entire system under a holistic reading of the prohibition.
If the system is deployed or remains in service with the social-scoring component active, the agency faces potential prohibition-level enforcement under Art. 5(1)(c), which carries the highest fine tier in the EU AI Act, and individual residents denied housing benefits could challenge decisions as void, triggering retroactive case review across all scored applicants.
Why this was triggered›
You indicated deployment in EU, the deployment is public sector, and your use-case description mentions social scoring.
The provision text›
Prohibits placing on the market, putting into service, or using AI systems for social scoring of natural persons by public authorities (or on their behalf) where the scoring leads to detrimental or unfavourable treatment in social contexts unrelated to the data's original collection, or where the treatment is unjustified or disproportionate to the underlying behaviour.
Where this is broadly unsettled›
Open interpretive points in the corpus, not yet tied to a specific launch.
- The line between prohibited public-authority social scoring and lawful credit/insurance risk modelling is contested where private deployers act under public-authority contracts.
- Whether private-sector deployers acting on behalf of public authorities fall within the prohibition depends on procurement structure and contractual control.
AI laws that may apply
2 surfaced across 1 lensGrouped by legal lens. Click any provision to see how it applies to this launch specifically.
other
2Requires certain deployers of high-risk AI systems to perform a Fundamental Rights Impact Assessment (FRIA) before first use. Applies to public bodies, private entities providing public services, and deployers of high-ri…
Grants data subjects the right not to be subject to a decision based solely on automated processing (including profiling) which produces legal effects or similarly significantly affects them. Three exceptions: contract n…
Not legal advice. Structured analysis of what a thoughtful counsel would consider given the inputs above. Does not substitute for counsel review or certify compliance.