- Eight out of ten public officials surveyed already use generative AI, and nearly half use it frequently in their daily work.
- Its use is concentrated in tasks such as document drafting and data analysis, with applications in citizen services still limited.
- To turn these individual advances into improvements for the public sector, clear guidelines, practical skills, and opportunities for institutional learning are needed.
Generative AI is used on a daily basis, but how is it being used in the public sector? An exploratory survey conducted by the Inter-American Development Bank (IDB) on the use of generative AI among 3,193 public officials from 19 countries in Latin America and the Caribbean provides some insights. Although the sample is not representative of the entire regional public sector, the results reveal a significant pattern among respondents: individual use of generative AI appears to be advancing faster than the rules, processes, and institutional capabilities needed to guide it.
85% of respondents in the public sector say they use generative AI tools in their work—whether frequently, occasionally, or by leading adoption initiatives. Nearly half say they use them frequently (see Figure 1).
Among those surveyed, the results show that there is already interest in experimenting with generative AI, even in the absence of an institutional strategy or formal mandate to guide its use. Without such support, experimentation may remain fragmented, progress unevenly, and unfold without the necessary guidance.
For now, generative AI is focused primarily on individual productivity tasks. Seventy-nine percent of respondents use it to draft documents, and 67% use it to analyze data. In contrast, only 10.5% use it for public service activities (see Figure 2).
Usage also varies by job type. Frequent use is most common among those in leadership positions or with coordination or policy-making roles, while occasional use is most common among those in administrative roles. In the latter group, 44% report using generative AI occasionally, compared to 32% who use it frequently. In addition, 39% say they do not know how to use it, which points to a skills gap in leveraging these tools at work.
The results suggest that adoption has focused primarily on individual productivity tasks. However, incorporating generative AI into these activities does not, in and of itself, transform the way an institution operates. For this technology to help improve public sector outcomes, a systemic overhaul is required: reviewing processes, strengthening data, clarifying responsibilities, and aligning its use with specific needs and outcomes.
There is a gap between reported individual use and perceptions of its adoption within teams. While 43% of respondents say they use generative AI frequently, only 19% believe that most of their colleagues actively use it—a difference of 24 percentage points (see Figure 3).
This gap does not mean other public officials are not using generative AI, but it does suggest that some experimentation may be happening with limited visibility. When organizations do not widely share lessons learned, they miss opportunities to identify useful practices, avoid repeating mistakes, and manage risks. Individual experience does not automatically translate into organizational knowledge.
This individual dimension matters more than it may seem. Adopting institutional tools powered by generative AI may be easier when public officials are already comfortable using these technologies and can champion change within their teams. Similarly, individual uncertainty about the technology may translate into resistance, low adoption, or superficial use of institutional solutions. For this reason, turning individual experimentation into collective learning is a key condition for managing organizational change.
For the civil servants surveyed, the challenge does not seem to be limited to adopting generative AI, but also to knowing how to use it wisely. Among those who use it occasionally, 44% admit that they do not know how to use it effectively.
Among those surveyed, 40% are concerned about potential errors or biases, and 36% are concerned about data privacy and security. In addition, nearly two out of every three identify cost as a barrier to accessing tools that are more advanced than the free options.
The lack of institutional guidance exacerbates these challenges. 82% are unclear about their organization’s stance: nearly half say that AI is permitted without clear guidelines, and more than a third are unaware of the rules (see Figure 4). This ambiguity can lead to unsafe uses or discourage legitimate applications. Without greater clarity, governments run the risk that adoption will remain fragmented and that they will fail to fully capitalize on some of its benefits.
This lack of clarity is consistent with a broader institutional gap documented by the Latin American Artificial Intelligence Index (ILIA 2025). Of the 19 countries analyzed, only nine have national AI strategies, and few have translated them into budgets or effective implementation mechanisms; another seven lack a consolidated roadmap. Although both studies measure different aspects, their findings point in the same direction: the adoption of AI appears to be advancing faster than the institutional capacity to guide and implement it.
1. Clear Rules Proportional to the Risk.
The absence of a clear policy creates uncertainty and could result in costs for the government. Officials need to know which uses are permitted, what information they must protect, and what level of review each task requires. The rules must protect organizations and citizens without blocking low-risk applications.
2. Judgment, Not Just Technical Skills.
Training must go beyond learning how to formulate better instructions for generative AI, known as prompts. It must also teach participants how to verify results, protect sensitive information, recognize biases, and decide when it is not appropriate to use this technology.
3. Organizational Learning.
Institutions need opportunities to share experiences, document use cases, and transform individual learnings into organizational knowledge. They also need to identify staff members who, based on their practical experience using these tools, can serve as internal champions of change. To improve processes and services, it is essential to link technology to specific problems, clear accountability, and verifiable results.
The potential of generative AI in the public sector will depend not only on the capabilities of the models, but also on the available systems and institutional capacities. Technology alone will not resolve fragmented processes, poor-quality data, unclear responsibilities, or coordination issues.
The results suggest that some of this adoption is already taking place in ways that are not readily apparent. The challenge for institutions is to support this process, manage its risks, and turn individual initiative into better outcomes for citizens.
ImplementaLAC is the IDB’s platform for public servants in Latin America and the Caribbean who seek to strengthen their capacity to design, manage, and implement public policies. It offers free courses on digital transformation, artificial intelligence, and other topics relevant to public management, as well as opportunities to exchange experiences and learn from solutions implemented in other countries in the region.
If you want to deepen your understanding of artificial intelligence and learn how to use it responsibly in your work, join ImplementaLAC. Gain access to resources and experiences that can help you turn your individual use of this technology into skills that strengthen public institutions.
Methodological Note: The exploratory survey was conducted between July and August 2026 and distributed to ImplementaLAC participants and users, as well as to contacts from other institutional networks. A total of 4,044 responses were received, of which 851 were excluded because they came from individuals who were not public officials or who did not provide complete information. The results presented are based on the remaining 3,193 responses, from 19 countries. The sample is not representative of the overall population of public officials in Latin America and the Caribbean and may reflect a self-selection bias toward individuals with a greater interest in innovation and digital tools. Therefore, the results allow us to identify patterns among the respondents but do not allow us to estimate their prevalence across the entire region.