Urban AI needs local oversight, not one-size-fits-all national rules

AI systems are no longer limited to experimental smart city projects. They are increasingly used in citizen services, transport analytics, planning compliance, environmental management, welfare-related assessments and service delivery. However, new research warns that municipalities remain poorly equipped to govern the technology at the point where its public effects are most immediate.

The study published in Smart Cities argues that cities should no longer be treated as passive implementers of national or corporate AI rules, but as frontline authorities responsible for turning broad ethical commitments into practical safeguards.

The research paper, titled Governing Urban AI from the Frontline: A Stage-Gate Framework for Municipal Algorithmic Decision-Making, proposes a stage-gate governance framework to help local governments assess, approve, monitor and revise AI systems before and after deployment, with special attention to accountability, fairness, public trust, institutional capacity and sustainability. The authors insist that municipal governments face a persistent gap between high-level AI principles and the real-world decisions involved in procuring, testing and using AI in public administration.

Cities face growing pressure as AI enters core public services

The same AI systems that help governments deliver core services can also introduce bias, strengthen surveillance, reduce transparency, widen digital inequality and shift power toward private technology vendors. The authors argue that this tension is especially acute in cities because municipal AI decisions are embedded in everyday life. Local governments manage services that residents experience directly, from urban planning and public health to local administration and infrastructure. A flawed AI system in these settings may not simply produce a technical failure. It can affect access to services, public confidence, civil rights and perceptions of democratic legitimacy.

Much of the current AI governance landscape is still shaped by national, supranational or corporate frameworks, the study claims. International principles such as those associated with the OECD, UNESCO and the EU AI Act provide useful norms on transparency, accountability, human rights and risk management. However, the authors argue that these frameworks often give limited direction on how municipalities should operationalize those principles under local constraints. Many city governments do not have large technical teams, formal AI ethics boards, advanced data systems or the budget flexibility needed to apply complex governance requirements without adaptation.

This gap leaves municipalities in a difficult situation. They are expected to deliver efficient, technology-enabled services, but they must also manage risks that may be difficult to detect before deployment. They often depend on third-party vendors for AI tools, raising concerns over data ownership, algorithmic control, procurement safeguards, vendor lock-in and explainability. Without clear municipal procedures, AI adoption can become fragmented across departments, driven by short-term efficiency targets or external pressure rather than public value.

The research identifies several governance challenges that municipalities must address together rather than separately. Institutional capacity is uneven, with some councils lacking dedicated expertise or permanent governance structures. Ethical oversight can be weak where there are no formal review bodies or operational toolkits. Data governance is often complicated by fragmented systems, poor interoperability, privacy risks and underinvestment in cybersecurity. Procurement processes may fail to demand adequate transparency from vendors. Public participation is often limited, even though AI systems can affect vulnerable or marginalized communities.

Responsible urban AI requires a local governance ecosystem that combines technical infrastructure, clear processes, policy frameworks, capacity building and public engagement. It should include secure data systems, defined workflows, regular algorithm checks, incident response procedures, procurement standards, audit mechanisms, staff training and channels for citizen feedback. Rather than treating AI governance as a one-time compliance exercise, the paper calls for continuous review and adaptation as technologies, regulations and community expectations change.

Stage-gate model turns broad AI principles into decision checkpoints

The study introduces a stage-gate framework designed for municipal algorithmic decision-making. The approach draws from project and innovation management, where complex initiatives are divided into stages separated by formal review gates. At each gate, decision-makers assess whether a project should proceed, pause, change direction or stop. The authors adapt this model for local government AI by embedding governance, risk assessment, ethics and stakeholder engagement across the AI lifecycle.

The framework starts with organizational groundwork. Before a city deploys or procures an AI system, it should establish enabling conditions, appoint a multidisciplinary team, define roles, clarify decision authority and build a shared understanding of AI opportunities and risks. This early stage is intended to prevent AI projects from advancing without basic readiness, internal coordination or accountability.

The next stage requires municipalities to identify the task clearly and determine whether AI is suitable at all. The authors stress that not every administrative problem requires an AI solution. Initial screening should test whether the proposed use case aligns with public value, whether simpler alternatives exist, whether relevant data are available and whether expected benefits justify the risks. This step is important because public-sector AI adoption can be driven by hype, political pressure or vendor promotion rather than a demonstrated need.

This is followed by ethical screening. At this point, the framework calls for local governments to embed principles such as fairness, transparency, accountability, privacy and inclusion into policies, procurement and evaluation. The study emphasizes that ethics should not be added after an AI tool has already been built or purchased. It should shape early decisions about design, data use, vendor requirements, public engagement and impact assessment.

A detailed investigation stage then examines financial planning, institutional capacity and data governance. Municipalities are urged to assess lifecycle costs, funding needs, external support, staff capacity, partnerships, data quality, consent rules, infrastructure readiness and open data opportunities. This is especially important for smaller councils that may not be able to maintain AI systems independently or absorb the costs of long-term vendor dependence.

The framework also requires a delivery model decision. Local governments must determine whether an AI system should be built in-house, outsourced, co-developed with partners or procured through a shared service arrangement. Each option has implications for transparency, cost, oversight, data protection and long-term control. The paper warns that delivery choices can shape the balance of power between municipalities and technology providers.

A later stage focuses on innovation and testing. The authors recommend safe experimentation through controlled environments, co-design, feedback channels and civic innovation practices. These steps allow municipalities to test systems before large-scale rollout, gather user input and adjust tools in response to evidence.

Risk assessment is one of the most critical gates. Local governments should classify risks, require algorithmic impact assessments, protect privacy, create complaints and redress mechanisms, and restrict or prohibit high-risk applications when necessary. The paper points to sensitive areas such as surveillance, predictive policing, biometric identification and welfare-related decision-making as areas where heightened scrutiny may be needed.

The framework then moves to policy assessment and governance integration. AI rules must be embedded into legal, procurement, data and accountability frameworks. Municipalities should update policies, include AI clauses in contracts, mandate transparency and ensure that local codes reflect community values as well as broader legal standards.

Implementation is treated as iterative rather than final. Municipalities should pilot systems, monitor performance and ethics, collect citizen feedback, conduct audits and define decommissioning criteria. This is followed by evaluation, where governments assess whether the AI system met its objectives, whether it caused unintended harm and whether it should be scaled, redesigned or retired.

The study also recognizes that not every municipality can implement all stages in full. For resource-constrained local governments, the authors propose a minimum viable pathway focused on five core checkpoints: enabling conditions, initial screening, ethical screening, delivery model decision-making and risk assessment. Monitoring and evaluation can be simplified or handled through shared regional arrangements. This makes the framework more adaptable to smaller councils while preserving the essential safeguards needed for responsible AI.

Public trust, procurement and capacity become decisive tests

AI systems used by local governments can affect who receives services, how risks are flagged, how neighborhoods are planned and how public resources are distributed. If residents cannot understand or challenge algorithmic decisions, public trust may weaken even when systems appear efficient.

The authors highlight public engagement as a core requirement. Participatory workshops, citizen forums, advisory panels and feedback mechanisms can help ensure that AI systems reflect community values and do not impose disproportionate burdens on marginalized groups. Public involvement is also important for surfacing concerns about labor impacts, service quality, data rights and surveillance before systems are normalized in government operations.

Procurement is another decisive test. Many local governments rely on vendors for AI systems, but procurement contracts may not always require explainability, audit access, data stewardship, bias testing or clear rules for system failure. The study argues that ethical and technical safeguards must be built into procurement from the outset. Without these protections, municipalities may adopt systems that are difficult to inspect, hard to modify and costly to replace.

The paper also calls for inter-local collaboration. Smaller or resource-limited municipalities may benefit from shared procurement platforms, peer networks, regional review bodies and common standards. Such collaboration can reduce duplication, improve bargaining power with vendors and help councils build capacity without each city having to create its own full AI governance unit.

Environmental and sustainability concerns also enter the analysis. AI systems can support smart city goals, including efficiency and better resource allocation, but they also carry material costs through energy-intensive computation, data infrastructure and electronic waste. The authors argue that municipal AI governance should consider social and ecological impacts, not just technical performance or administrative speed.

The study's limitations can't be ignored. It is a conceptual and exploratory paper, not a field-tested evaluation of the proposed framework. The authors draw on literature, policy analysis and documented municipal practices, but they do not claim that the model has been validated through direct implementation. They call for future research using comparative case studies, pilot programs, participatory action research and longitudinal analysis to test how the framework works across different municipal contexts.

In a nutshell, responsible AI cannot be delivered through abstract principles alone. It requires institutional readiness, public participation, procurement discipline, risk review, monitoring and the political will to treat residents as rights-bearing citizens rather than data points in automated systems. The research positions municipalities not as the weakest link in AI governance, but as the level of government best placed to align urban innovation with accountability, equity and public value.