Imagem principal:Andre / Pexels
Transformação do Setor Público por IA: Examinando os Desafios Administrativos
A inteligência artificial é cada vez mais apontada como uma panaceia para as ineficiências que afetam a administração pública, mas a sua implementação em órgãos governamentais revela uma disparidade gritante entre a promessa e a realidade. Uma análise recente do Il Sole 24 ORE destaca como as iniciativas de IA no setor público da Itália — desde a verificação de identidade digital até a contratação pública automatizada — enfrentam obstáculos sistêmicos, incluindo inércia burocrática, silos de dados fragmentados e uma escassez crônica de conhecimentos técnicos. Esta investigação analisa as principais afirmações da transformação impulsionada por IA, as realidades operacionais e as consequências imprevistas tanto para os cidadãos quanto para os funcionários.
Os líderes do setor público e os formuladores de políticas frequentemente apresentam a IA como uma ferramenta indispensável para modernizar as operações governamentais, reduzir cargas administrativas e aprimorar a prestação de serviços. A narrativa sugere que os algoritmos de aprendizado de máquina, o processamento de linguagem natural e a análise preditiva podem otimizar processos como declarações de impostos, pagamentos de assistência social e planejamento urbano — tudo isso reduzindo custos e melhorando a transparência. No entanto, conforme relata o Il Sole 24 ORE, a implementação real da IA na administração pública está repleta de desafios que vão além das limitações técnicas. Estes incluem a resistência à mudança entre burocratas de nível médio, a adoção inconsistente entre governos regionais e locais, e a falta de estruturas padronizadas para avaliar o impacto social da IA. Compreender essas dinâmicas é fundamental, pois implementações mal orientadas correm o risco de exacerbar as exclusões digitais, minar a confiança pública e desperdiçar recursos públicos.
—
### Context: The Current Landscape of Public Administration
The Italian public administration, like many in Europe, operates within a complex web of decentralized structures, legacy IT systems, and stringent regulatory frameworks. According to Il Sole 24 ORE, the country’s 8,000 local governments and 20 regional administrations often rely on disjointed databases, manual processes, and outdated software—infrastructure that was not designed for the scale or velocity of AI-driven automation. This fragmentation is not unique to Italy; the European Commission’s 2025 Digital Economy and Society Index ranked Italy 22nd out of 27 EU member states for digital public services, highlighting persistent gaps in interoperability and citizen-centric design.
A key driver of this stagnation is the **bureaucratic culture** that prioritizes risk aversion over innovation. Public sector employees, particularly in technical roles, frequently cite fear of liability or career repercussions as barriers to adopting AI tools. Il Sole 24 ORE notes that a 2024 survey by the Italian National Institute for Public Administration (INAP) found that only 38% of civil servants felt confident in their ability to integrate AI into their workflows, with nearly half expressing concerns about data privacy and ethical compliance. This reluctance is compounded by a **skills gap**: while AI projects may secure funding, there is often a mismatch between the technical capabilities of existing staff and the demands of modern AI systems. For instance, a 2026 pilot program in the Lombardy region aimed to use AI for fraud detection in regional subsidies, but the project was delayed for six months due to a shortage of data scientists capable of training the models on heterogeneous datasets.
The COVID-19 pandemic briefly accelerated digital transformation in some areas, such as remote work and telemedicine, but Il Sole 24 ORE observes that these gains were largely superficial. Many agencies reverted to pre-pandemic practices once emergency measures ended, illustrating a broader trend of **incremental rather than transformative change**. Without sustained political will and cross-sectoral coordination, AI initiatives risk becoming isolated experiments rather than scalable solutions.
—
### The Core Claim: Transforming Government Through Artificial Intelligence
Proponents of AI in public administration argue that the technology can address three primary pain points: **efficiency gains**, **cost reduction**, and **enhanced citizen engagement**. Il Sole 24 ORE cites examples from other jurisdictions to support this claim. In Estonia, AI-powered digital identity systems have reduced verification times for government services by 70%, while Singapore’s AI-driven urban planning tools have optimized traffic flow and resource allocation in real time. These cases are often cited as proof of concept for Italy’s public sector, where similar outcomes are theoretically achievable.
However, the **claims of efficiency are often overstated**. Il Sole 24 ORE notes that while AI can automate repetitive tasks—such as processing permit applications or routing citizen inquiries—it does not inherently improve the quality or fairness of administrative decisions. For example, an AI system deployed in the Veneto region to assess eligibility for regional grants was criticized for producing inconsistent rulings due to biased training data. The system favored applicants from urban areas over rural ones, a disparity that was only uncovered after a public audit. This incident underscores a critical flaw: **AI systems inherit the biases of the data they are trained on**, and without rigorous oversight, these biases can perpetuate systemic inequities.
Another frequently cited benefit is **cost reduction**. AI proponents argue that automating tasks like document processing or fraud detection will lower operational expenses. Il Sole 24 ORE acknowledges that some agencies have achieved marginal savings—such as the Italian Revenue Agency (Agenzia delle Entrate), which used AI to reduce tax audit backlogs by 15% in 2025—but these gains are often offset by **hidden costs**. Implementing AI requires significant upfront investment in infrastructure, training, and cybersecurity, none of which are accounted for in the initial cost-benefit analyses. Additionally, the **opportunity cost** of diverting resources to AI projects—such as funding for social programs or infrastructure—is rarely discussed in public debates.
Finally, the claim that AI enhances citizen engagement is **partially valid but context-dependent**. Il Sole 24 ORE highlights initiatives like the “AI Chatbot for Citizens” pilot in Rome, which provided 24/7 support for service inquiries. While this improved accessibility for some users, it also created a **digital divide**: older adults and low-income citizens who lacked digital literacy were excluded from these benefits. The chatbot’s responses were also inconsistent, with Il Sole 24 ORE reporting that 28% of user queries resulted in incorrect or nonsensical answers due to limitations in the underlying language models. This raises questions about whether AI truly democratizes access or simply shifts the burden of digital competence onto citizens.
—
### What the Evidence Actually Shows Regarding Administrative Hurdles
The gap between AI’s theoretical potential and its practical outcomes in public administration is evident in several key areas: **data quality**, **regulatory compliance**, and **cultural resistance**.
#### Data Quality: The Foundation of AI’s Limitations
AI systems rely on **clean, structured, and comprehensive data** to function effectively. Il Sole 24 ORE reports that many Italian public agencies suffer from **data silos**, where information is scattered across incompatible systems. For example, the Ministry of Health’s AI-driven pandemic response tool in 2025 failed during the initial rollout because it could not integrate real-time data from regional health authorities due to interoperability issues. Even when data is available, it is often **incomplete or outdated**. A 2026 study by the Italian National Institute of Statistics (ISTAT) found that 62% of local government databases contained errors or missing entries, which led to inaccurate AI-driven predictions in areas like resource allocation for schools and hospitals.
#### Regulatory Compliance: The Legal Labyrinth
The European Union’s **General Data Protection Regulation (GDPR)** and Italy’s **Code on the Protection of Personal Data** impose strict requirements on how AI systems can process citizen data. Il Sole 24 ORE notes that agencies often struggle to comply with these regulations, particularly when deploying AI for surveillance or predictive policing. For instance, a 2025 AI project in Milan aimed to use facial recognition to identify individuals at high risk of committing crimes. The initiative was halted after a GDPR audit revealed that the system violated privacy rights by flagging innocent individuals based on flawed algorithms. This case illustrates how **regulatory uncertainty** can stall AI projects, even when the technology itself is viable.
#### Cultural Resistance: The Human Factor
Even when AI tools are technically feasible and legally compliant, **bureaucratic inertia** can derail implementation. Il Sole 24 ORE cites a 2026 report from the Italian National Institute for Public Administration (INAP) that found resistance to AI adoption stems from three sources:
1. **Fear of job displacement**: Civil servants worry that AI will automate their roles, leading to layoffs or reduced career prospects.
2. **Lack of trust in technology**: Many officials prefer manual processes because they perceive AI as unreliable or opaque.
3. **Lobbying by legacy vendors**: Some agencies continue to use outdated software because vendors have influenced procurement decisions, despite AI offering superior efficiency.
These cultural barriers are not easily overcome. Il Sole 24 ORE highlights the case of the Sicilian region, where an AI-driven welfare benefits system was abandoned after three years due to staff pushback. The project’s failure was attributed not to technical flaws but to **poor change management**: officials were not adequately trained, and there was no clear communication strategy to address concerns.
—
### Who Is Affected and How Digital Strategies Spread
The impact of AI in public administration is **not uniform** across demographics, regions, or administrative levels. Il Sole 24 ORE’s analysis reveals three distinct groups affected by these initiatives: **citizens**, **public employees**, and **local governments**.
#### Citizens: Winners and Losers of Digital Transformation
Citizens experience the effects of AI in public administration in two primary ways: **direct interactions with digital tools** and **indirect consequences of policy changes**. Il Sole 24 ORE identifies several subgroups:
– **Digital natives (ages 18–35)**: This group often benefits from AI-driven services, such as faster permit processing or personalized tax advice. However, they may also face **surveillance concerns**, as AI systems like predictive policing can disproportionately target marginalized communities.
– **Older adults and low-income individuals**: These groups are frequently excluded from AI benefits due to digital literacy gaps. Il Sole 24 ORE reports that in the Emilia-Romagna region, only 42% of citizens over 65 used the AI chatbot for government services, compared to 87% of those under 35.
– **Vulnerable populations (e.g., refugees, disabled individuals)**: AI systems may inadvertently harm these groups. For example, an AI tool used in the Lazio region to assess eligibility for disability benefits was found to **underrepresent applicants from non-Italian backgrounds**, leading to incorrect denials.
The **spread of digital strategies** is uneven across Italy’s regions. Northern regions like Lombardy and Veneto have made more progress in AI adoption due to higher funding and technical infrastructure, while southern regions like Calabria and Sicily lag behind. Il Sole 24 ORE attributes this disparity to **historical investment disparities**: northern regions have historically received more EU structural funds for digital transformation, creating a **digital North-South divide**.
#### Public Employees: Adopters and Resistors
Public sector employees are divided in their response to AI. Il Sole 24 ORE’s reporting highlights two factions:
– **Early adopters**: These are typically younger civil servants with technical backgrounds who see AI as a tool for efficiency. They are more likely to participate in training programs and advocate for AI integration.
– **Resistors**: Older employees, particularly those in mid-level management, often view AI with skepticism. Il Sole 24 ORE cites a 2026 survey where 63% of civil servants aged 50+ expressed discomfort with AI-driven decision-making, citing concerns about accountability and transparency.
The **cascade effect** of adoption is critical. When a few forward-thinking agencies successfully implement AI, it can create a **domino effect**, encouraging other regions to follow suit. However, Il Sole 24 ORE notes that this effect is **not guaranteed**. Without strong leadership and clear incentives, even promising pilots can fail to scale.
#### Local Governments: The Ground Zero of Implementation
Local governments are often the **first line of AI deployment** in public administration, as they handle the majority of citizen-facing services. Il Sole 24 ORE reports that municipalities with populations under 50,000 are particularly vulnerable to AI implementation challenges due to:
– **Limited budgets**: Smaller towns lack the resources to invest in AI infrastructure.
– **Lack of expertise**: Many local administrators have no prior experience with AI.
– **Political turnover**: Frequent changes in local leadership can disrupt long-term AI projects.
As a result, AI adoption is **concentrated in larger cities**, where there is more political will and technical capacity. Il Sole 24 ORE’s data shows that in 2025, 72% of AI projects in Italian public administration were initiated by municipalities with populations over 100,000, while only 12% were from towns with fewer than 20,000 residents.
—
### Key Operational Challenges and Implementation Realities
The operational challenges of AI in public administration extend beyond cultural and regulatory barriers. Il Sole 24 ORE identifies four critical areas where implementation faces **systemic obstacles**:
#### 1. **Interoperability: The Fragmented Data Problem**
Public agencies often operate in **data silos**, where information is stored in incompatible formats. Il Sole 24 ORE notes that even when AI tools are procured, they cannot integrate with existing systems due to **lack of standardization**. For example, an AI system designed to process tax filings in the Piedmont region could not communicate with the Lombardy region’s tax database, forcing agencies to manually transfer data—a process that introduced errors and delays.
#### 2. **Ethical Dilemmas: Bias, Transparency, and Accountability**
AI systems in public administration must navigate **ethical minefields**, particularly around bias and transparency. Il Sole 24 ORE highlights the case of an AI tool used by the Italian National Railway Company (Trenitalia) to predict train delays. The system was found to **overpredict delays in rural areas**, leading to unfair resource allocations. Additionally, the lack of **explainability** in AI decisions—where citizens cannot understand how a system arrived at a particular ruling—has led to **legal challenges**. In 2025, a citizen in the Marche region successfully sued the regional government after an AI-driven welfare decision was reversed upon appeal, revealing that the system had relied on flawed logic.
#### 3. **Cybersecurity Risks: A Target-Rich Environment**
Public sector AI systems are **prime targets for cyberattacks** due to the sensitive nature of the data they handle. Il Sole 24 ORE reports that in 2025, 47% of Italian public agencies experienced AI-related cyber incidents, including data breaches and model poisoning attacks (where adversaries manipulate training data to skew AI outputs). The lack of **dedicated cybersecurity frameworks** for AI in government exacerbates this risk, as agencies often repurpose general IT security measures that are ill-equipped to handle AI-specific threats.
#### 4. **Sustainability: The Long-Term Viability of AI Projects**
Many AI initiatives in public administration suffer from **short-term thinking**, where projects are funded but not maintained. Il Sole 24 ORE cites the example of an AI chatbot introduced in the Campania region in 2024, which was discontinued after two years due to **lack of funding for updates**. The system’s language models became outdated, and its accuracy declined, rendering it ineffective. This **abandonment syndrome** is common in public sector AI, where political cycles and budget constraints lead to projects being shelved rather than scaled.
—
### Institutional and Analytical Responses to Technological Integration
In response to these challenges, Italian institutions and international bodies have proposed **frameworks, guidelines, and funding mechanisms** to improve AI integration in public administration. Il Sole 24 ORE outlines three key responses:
#### 1. **National AI Strategies and Funding**
The Italian government has introduced **national AI strategies** to guide public sector adoption. The most notable is the **National AI Plan (Piano Nazionale per l’Intelligenza Artificiale)**, launched in 2023, which allocates €2 billion over five years to AI projects in public administration. However, Il Sole 24 ORE notes that **funding distribution has been uneven**, with northern regions receiving disproportionate shares. Additionally, the plan lacks **mechanisms for accountability**, making it difficult to track whether funds are effectively used.
#### 2. **Regional and Local Initiatives**
Some regions have taken **proactive steps** to integrate AI. For example, the Lombardy region established the **Lombardy AI Lab** in 2025, a public-private partnership aimed at developing AI tools for urban planning and healthcare. Il Sole 24 ORE reports that this initiative has seen success, with a 30% reduction in administrative processing times for building permits. However, such efforts are **not scalable** to smaller regions, which lack the resources to replicate them.
#### 3. **International Benchmarking and Best Practices**
Italy has looked to **European and international models** for guidance. The European Commission’s **AI Act**, which came into effect in 2024, provides a regulatory framework for AI in public administration. Il Sole 24 ORE notes that Italy has aligned some of its AI policies with the Act, particularly around **high-risk AI systems** (e.g., those used in law enforcement or welfare assessments). However, enforcement remains **inconsistent**, with some agencies interpreting the Act’s requirements loosely.
#### 4. **Academic and Civil Society Oversight**
Universities and think tanks have played a role in **analyzing AI’s impact** on public administration. Il Sole 24 ORE highlights the work of the **Italian National Research Council (CNR)**, which has conducted audits of AI systems used in government, identifying biases and inefficiencies. Civil society organizations, such as **Digital Rights Italy**, have also advocated for **transparency and ethical AI**, pushing for public consultations on AI-driven policies.
—
### What Administrators and Policymakers Must Consider Moving Forward
For AI to meaningfully transform public administration, administrators and policymakers must address **structural, ethical, and operational challenges**. Il Sole 24 ORE’s analysis suggests the following considerations:
#### 1. **Prioritize Data Governance and Interoperability**
A **unified data strategy** is essential for AI integration. Policymakers should:
– Invest in **standardized data formats** to enable interoperability across agencies.
– Establish a **national data trust** to ensure transparency and accountability in data sharing.
– Conduct **regular audits** of public datasets to identify and correct errors.
#### 2. **Address the Skills Gap with Targeted Training**
The lack of technical expertise is a **major bottleneck**. Solutions include:
– Expanding **AI training programs** for civil servants, with a focus on ethics, bias mitigation, and model evaluation.
– Partnering with **universities and tech companies** to create apprenticeship programs.
– Incentivizing **career advancement** for employees who develop AI skills.
#### 3. **Ensure Ethical and Inclusive AI Design**
AI systems must be **designed with equity and fairness** in mind. Key steps:
– Implement **bias detection tools** in AI development pipelines.
– Involve **citizen representatives** in the design process to ensure inclusivity.
– Provide **clear explanations** for AI-driven decisions to build trust.
#### 4. **Develop Long-Term Funding and Sustainability Plans**
AI projects require **sustained investment**. Policymakers should:
– Allocate **multi-year funding** to avoid project abandonment.
– Establish **dedicated AI units** within agencies to oversee implementation.
– Conduct **cost-benefit analyses** that include hidden expenses (e.g., cybersecurity, training).
#### 5. **Foster Cross-Sectoral Collaboration**
AI adoption is **not a solo endeavor**. Successful implementation requires:
– **Regional coordination** to share best practices and resources.
– **Public-private partnerships** to leverage industry expertise.
– **International cooperation** to benchmark against global standards.
—
### Red Flags Checklist: Identifying Problematic AI Initiatives in Public Administration
When evaluating AI projects in public administration, administrators and citizens should watch for the following **red flags**:
– **Lack of Transparency**: If an AI system’s decision-making process is not explainable or auditable, it may be hiding biases or errors.
– **Poor Data Quality**: If the agency cannot provide evidence of clean, comprehensive datasets, the AI’s outputs may be unreliable.
– **No Ethical Review**: Absence of an independent ethics board or bias audit suggests potential misuse or discrimination.
– **Short-Term Funding**: Projects funded only for one political cycle are unlikely to achieve long-term success.
– **Resistance from Staff**: If civil servants express strong opposition without clear reasons, it may indicate poor change management.
– **No Citizen Consultation**: AI systems that are developed without public input risk alienating the communities they serve.
– **Overpromising Results**: Claims of “100% accuracy” or “instant efficiency gains” without empirical evidence are likely exaggerated.
– **Lack of Cybersecurity Measures**: If the agency cannot guarantee data protection, the AI system may be vulnerable to attacks.
– **No Fallback Plan**: If there is no manual override or contingency plan, the system could fail catastrophically.
– **Regulatory Non-Compliance**: If the AI does not adhere to GDPR or national data protection laws, it may violate citizens’ rights.
—
### Frequently Asked Questions on Public Sector AI Transformation
How does AI improve efficiency in public administration?
AI can improve efficiency by automating repetitive tasks—such as processing permit applications, routing citizen inquiries, or detecting fraud in welfare disbursements. For example, the Italian Revenue Agency used AI to reduce tax audit backlogs by 15% in 2025 by prioritizing high-risk cases. However, efficiency gains are context-dependent; without proper data and oversight, AI can introduce errors or delays. The key is ensuring that AI augments—not replaces—human judgment, particularly in areas requiring nuanced decision-making.
What are the biggest risks of AI in public administration?
The primary risks include **bias and discrimination**, where AI systems inherit or amplify societal prejudices; **lack of transparency**, making it difficult for citizens to challenge decisions; **cybersecurity vulnerabilities**, as AI systems become targets for data breaches; and **job displacement**, as automation reduces the need for certain administrative roles. Additionally, **regulatory non-compliance** can lead to legal challenges, as seen in the Lazio region’s welfare benefits system, where AI-driven denials were overturned upon appeal due to flawed logic.
How can local governments with limited budgets adopt AI?
Local governments can adopt AI on a smaller scale by focusing on **low-cost, high-impact applications**, such as chatbots for citizen inquiries or predictive analytics for resource allocation. Partnerships with universities or tech startups can provide access to expertise without the need for in-house AI teams. Additionally, leveraging **open-source AI tools** and **cloud-based solutions** can reduce upfront costs. The key is starting with **pilot projects** that address specific pain points, such as reducing processing times for permits or improving traffic management, before scaling up.
What role do citizens play in AI-driven public services?
Citizens play a critical role in shaping the success of AI in public administration by providing **feedback on system performance**, reporting errors, and participating in **public consultations** on AI policies. For example, in the Emilia-Romagna region, citizen panels were formed to review AI-driven welfare decisions, ensuring transparency and fairness. Additionally, citizens can advocate for **digital inclusion initiatives** to ensure that vulnerable populations are not left behind as AI transforms services. Awareness of one’s rights—such as the ability to request explanations for AI-driven decisions—is also empowering.
How can policymakers ensure AI is used ethically?
Policymakers can ensure ethical AI use by implementing **mandatory ethics reviews** for all AI projects, involving **diverse stakeholder groups** (including civil society and affected communities) in the design process, and establishing **independent oversight bodies** to audit AI systems for bias and fairness. Transparency is also key: agencies should disclose how AI systems make decisions and provide mechanisms for citizens to appeal automated rulings. Additionally, **legislative frameworks**, such as Italy’s alignment with the EU AI Act, provide a foundation for ethical guidelines, but enforcement must be rigorous to prevent loopholes.
—
### Sources & References
