Where AI Meets the Town Hall: A Basque Study Finds That Local Governments Are Unequally Prepared for Algorithmic Risk
Artificial intelligence is moving rapidly into public services, from automated benefit assessments and digital identity systems to predictive tools used in health, transport and social care. Yet a new study from Gipuzkoa in the Basque Country suggests that the most important question may not be whether governments can deploy AI, but whether they possess the institutional capacity to understand and control its consequences. Action research involving six civil society organisations, seven provincial departments and 11 municipalities found that the ability to anticipate AI-related risks is unevenly distributed across local governance. The researchers argue that responsible AI must be built around territorial digital inclusion: ensuring that people can access services, appear fairly in public datasets and participate in decisions that increasingly affect their lives.
Published in Law, Ethics & Technology, the study examines an action-research programme conducted through the Provincial Council of Gipuzkoa between 2024 and 2026. Rather than evaluating the accuracy of a particular algorithm, the researchers investigated how public institutions learn to govern emerging technologies before those technologies become deeply embedded in administrative systems. The programme brought together public officials, municipalities, researchers, policymakers and civil society groups representing vulnerable communities. Its work was organised through an eight-step Science-for-Policy roadmap combining knowledge exchange, stakeholder engagement, evidence development and open-science dissemination. This approach treats AI governance as an ongoing policy capability rather than a one-time compliance exercise.
The researchers found that different actors identify different kinds of risk. Civil society organisations often detect problems through direct experience with exclusion, discrimination and barriers to digital access. Their knowledge may reveal how a system affects people who depend on mobile phones, lack stable internet connections, have disabilities or struggle to navigate official platforms. Provincial departments and directorates are more likely to convert these concerns into formal mechanisms, including algorithmic audits, data-governance structures, oversight procedures and accountability frameworks. Municipalities, particularly those with fewer financial and technical resources, face a different reality. They may have responsibility for delivering services but lack specialists, data infrastructure or clear authority over digital-rights protection.
That uneven distribution of expertise can produce what the authors describe as a territorial governance gap. An AI system designed according to national or regional standards may operate very differently in a small municipality than in a larger administration with dedicated legal, technical and ethical teams. Local authorities can struggle to determine who is accountable when an automated decision causes harm, whether affected residents have a meaningful route to challenge it, and how personal data are being collected or reused. The study also identified uncertainty over who is responsible for digital inclusion itself. Without clear ownership, governments may introduce online services while overlooking residents who cannot reliably access them or who require human assistance.
The Gipuzkoa case highlights how technical systems can reproduce social inequalities through seemingly neutral data. If vulnerable groups are missing from administrative datasets, an AI model trained on those datasets may treat their needs as statistically insignificant. If digital access depends on mobile devices, residents with limited connectivity may be less visible to the system than those who interact frequently online. Algorithmic bias does not necessarily begin with an intentionally discriminatory code. It can emerge from incomplete records, unequal participation, poorly defined categories or decisions made without consulting the communities affected. For this reason, the researchers link data quality to democratic participation, arguing that people excluded from the data-generation process may also be excluded from the policies built on that data.
“The key question for public administrations is not simply whether to adopt AI, but whether they have the institutional capacity to govern it democratically,” said lead author Igor Calzada of the University of the Basque Country and Ikerbasque. He argues that foresight must be connected to digital inclusion, human rights and public participation from the beginning of the policy process. In practical terms, this means asking who might be harmed before an automated system is procured, identifying which groups are absent from the evidence base, and creating channels through which residents can influence design and implementation. It also means treating explainability not merely as a technical feature, but as a condition for public accountability.
The study proposes a set of governance principles designed to help administrations build that capacity. They include foresight, transparency and explainability, participation and co-creation, human oversight, accountability, ethical and human-centred AI, and protection of digital rights. These principles are intended to work together. Transparency may reveal how a system is used, but participation can show whether its objectives reflect public needs. An audit may identify statistical disparities, while human oversight can determine whether those disparities justify intervention. Accountability mechanisms are essential when responsibility is divided among software vendors, departments and local authorities. The researchers therefore describe anticipatory governance as a cycle of learning, coordination and revision rather than a fixed checklist.
Co-author Itziar Eizaguirre, of the Human Rights and Democratic Culture Directorate at the Provincial Council of Gipuzkoa, said that people experiencing digital exclusion often identify problems that administrative datasets cannot capture. Bringing civil society organisations, municipalities and provincial departments into the same process can make those experiences visible before policy decisions are finalised. This collaborative model also changes the role of public participation. Instead of asking residents to comment on a completed technology project, institutions can involve them in defining the problem, assessing possible harms and deciding what safeguards are necessary. In that sense, participation becomes part of the technical governance process, influencing what data are collected, what outcomes are measured and when automation should not be used.
The authors caution that Gipuzkoa should not be treated as a universal blueprint. Its fiscal autonomy and administrative resources create conditions that may not exist in other territories, especially in regions with smaller budgets or less developed digital infrastructure. The transferable lesson is not a single organisational structure but a set of adaptable practices: co-production with civil society, coordination across departments, cooperation between provincial and municipal authorities, and a rights-based approach to technology policy. The researchers also stress that their study examines governance processes rather than the long-term performance of specific AI systems. It does not yet show whether the proposed roadmap produces measurable improvements in public services or reduces documented discrimination.
Further comparative and longitudinal research will be needed to determine how anticipatory capacity develops in different political and economic settings. Such work could examine whether municipalities with limited resources can share technical expertise, whether civil society participation changes procurement decisions, and how oversight systems function after AI tools are deployed. For now, the Gipuzkoa study offers a warning with implications far beyond the Basque Country: AI may be introduced through software, but its social consequences are shaped by institutions, communities and the quality of democratic decision-making around it. The paper, “Anticipatory AI governance for territorial digital inclusion strategies: action research within the Provincial Council of Gipuzkoa (Basque Country),” was published in Law, Ethics & Technology.
Subject of Research: Not applicable
Article Title: Anticipatory AI governance for territorial digital inclusion strategies: action research within the Provincial Council of Gipuzkoa (Basque Country)
News Publication Date: 13-Jul-2026
Web References: https://doi.org/10.55092/let20260008
References: Calzada I, Eizaguirre I. “Anticipatory AI governance for territorial digital inclusion strategies: action research within the Provincial Council of Gipuzkoa (Basque Country).” Law, Ethics & Technology. 2026(3):0008.
Image Credits: Igor Calzada/University of the Basque Country (UPV/EHU); Itziar Eizaguirre/Provincial Council of Gipuzkoa
Keywords: Artificial intelligence governance, territorial digital inclusion, public services, algorithmic accountability, digital rights, public participation, municipalities, Science-for-Policy, anticipatory governance, Basque Country
