REVISTA EDUCACIÓN SUPERIOR Y SOCIEDAD

2025, Vol. 37 Nro. 2 (jul.- dic.), 288-307.

https://doi.org/10.54674/ess.v37i2.1278

e-ISSN: 2610-7759

Recibido 2026-03-31│Revisado 2026-04-14

Aceptado 2026-04-28│Publicado 2026-05-30

 

 

 

 

AI Adoption and Governance in Latin American and Caribbean Higher Education: Findings from a Regional Survey

Adopción y gobernanza de la IA en la educación superior de América Latina y el Caribe: resultados de una encuesta regional

 

Arianna Valentini Céspedes* @ https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcSNfWYTcCAZvEGsSFZ1txbWDB-BbGS9NXIvs46eBwRcKcb97noqr8ag9zTjvaHe_8qoX9A&usqp=CAU

* UNESCO International Institute for Higher Education in Latin America and the Caribbean, Caracas, Venezuela

 

 

ABSTRACT

This article presents the findings of a quantitative, descriptive regional survey conducted with 200 higher education institutions across 19 countries in Latin America and the Caribbean (LAC) between August and October 2025. Drawing on a theoretical framework grounded in the three-mission model of the university (teaching and learning, research, and community engagement), the study maps AI adoption across these three missions as well as administration as an additional analytical dimension. It examines governance as a fifth, cross-cutting dimension that shapes the conditions under which integration is taking place. The results reveal a clear hierarchy of adoption, with teaching and learning leading at 73.5%, followed by research (57.0%), administration (34.1%), and community engagement (20.0%). Governance conditions lag far behind; only 26.0% of institutions have a formal AI strategy, 18.5% have institution-wide policies, 8.0% have a dedicated AI budget, and just 9.0% have formal evaluation mechanisms. The study identifies a persistent public-private divide, a structural vulnerability in specialised universities, significant gaps in student protection, and a monitoring deficit that complicates institutional accountability.

KEYWORDS: Higher Education; Artificial Intelligence; Governance; Latin America and the Caribbean; AI Policy

 

Adopción y gobernanza de la IA en la educación superior de América Latina y el Caribe: resultados de una encuesta regional

RESUMEN

Este artículo presenta los resultados de una encuesta regional de diseño cuantitativo y descriptivo realizada a 200 instituciones de educación superior en 19 países de América Latina y el Caribe (ALC), entre agosto y octubre de 2025. A partir de un marco teórico fundamentado en el modelo de las tres misiones universitarias (docencia y aprendizaje, investigación y servicio a la comunidad), el estudio mapea la adopción de la IA en estas tres misiones, así como en la administración como dimensión analítica adicional, y examina la gobernanza como quinta dimensión transversal que determina las condiciones en que dicha integración tiene lugar. Los resultados revelan una jerarquía clara de adopción, con la docencia y el aprendizaje a la cabeza con un 73,5%, seguidos por la investigación (57,0%), la administración (34,1%) y el servicio a la comunidad (20,0%). Las condiciones de gobernanza quedan muy rezagadas: solo el 26,0% de las instituciones cuenta con una estrategia formal de IA, el 18,5% tiene políticas institucionales amplias, el 8,0% dispone de presupuesto específico para IA y apenas el 9,0% cuenta con mecanismos formales de evaluación. El estudio identifica una brecha público-privada persistente, una vulnerabilidad estructural en las universidades especializadas, importantes déficits en la protección del estudiantado y un déficit de seguimiento que dificulta la rendición de cuentas institucional.

PALABRAS CLAVE: educación superior; inteligencia artificial; gobernanza; América Latina y el Caribe; Política de IA

 

Adoção e Governação da Inteligência Artificial na Educação Superior da América Latina e do Caribe: Resultados de um Levantamento Regional

RESUMO

Este artigo apresenta os resultados de um levantamento regional de carácter quantitativo e descritivo, realizado junto de 200 instituições de educação superior em 19 países da América Latina e do Caribe (ALC), entre agosto e outubro de 2025. Com base num quadro teórico fundamentado no modelo das três missões universitárias (ensino e aprendizagem, investigação e serviço à comunidade), o estudo mapeia a adoção da IA nestas três missões, bem como na administração enquanto dimensão analítica adicional, e examina a governação como quinta dimensão transversal que determina as condições em que essa integração ocorre. Os resultados revelam uma hierarquia clara de adoção, com o ensino e a aprendizagem em primeiro lugar, com 73,5%, seguidos da investigação (57,0%), da administração (34,1%) e do serviço à comunidade (20,0%). As condições de governação ficam muito aquém: apenas 26,0% das instituições dispõem de uma estratégia formal de IA, 18,5% têm políticas institucionais abrangentes, 8,0% possuem um orçamento dedicado à IA e apenas 9,0% contam com mecanismos formais de avaliação. O estudo identifica uma persistente divisão entre instituições públicas e privadas, uma vulnerabilidade estrutural nas universidades especializadas, défices significativos na proteção dos estudantes e um défice de acompanhamento que dificulta a prestação de contas institucional.

KEY WORDS: Educação Superior; Inteligência Artificial; Governação; América Latina e Caribe; Políticas de IA

 

Adoption et Gouvernance de l'Intelligence Artificielle dans l'Enseignement Supérieur en Amérique Latine et dans les Caraïbes: Résultats d'une Enquête Régionale

RÉSUMÉ

Cet article présente les résultats d'une enquête régionale à caractère quantitatif et descriptif, menée auprès de 200 établissements d'enseignement supérieur dans 19 pays d'Amérique latine et des Caraïbes (ALC), entre août et octobre 2025. S'appuyant sur un cadre théorique ancré dans le modèle des trois missions universitaires (enseignement et apprentissage, recherche et service à la communauté), l'étude cartographie l'adoption de l'IA dans ces trois missions ainsi que dans l'administration en tant que dimension analytique supplémentaire, et examine la gouvernance comme cinquième dimension transversale déterminant les conditions dans lesquelles cette intégration s'opère. Les résultats révèlent une hiérarchie claire d’adoption: l'enseignement et l'apprentissage arrivent en tête avec 73,5 %, suivis de la recherche (57,0 %), de l'administration (34,1 %) et du service à la communauté (20,0 %). Les conditions de gouvernance accusent un retard considérable: seulement 26,0 % des établissements disposent d'une stratégie formelle en matière d'IA, 18,5 % ont des politiques à l'échelle de l'institution, 8,0 % disposent d'un budget dédié à l'IA et seulement 9,0 % ont mis en place des mécanismes formels d'évaluation. L'étude identifie une fracture persistante entre établissements publics et privés, une vulnérabilité structurelle dans les universités spécialisées, des lacunes importantes dans la protection des étudiants et un déficit de suivi qui rend difficile la responsabilisation institutionnelle.

MOTS-CLÉS: Enseignement supérieur; Intelligence artificielle; Gouvernance; Amérique latine et Caraïbes; Politique en matière d'IA

 

 

1. INTRODUCTION

The integration of artificial intelligence (AI) into higher education has accelerated, generating both transformative opportunities and urgent governance challenges. Globally, 56 countries had adopted national AI strategies as of 2024, a majority of which underline the critical role of higher education in AI education, research, and innovation, and 55% had already launched AI graduate programmes to encourage highly skilled professionals in the field (UNESCO IESALC, 2025b). Yet institutional readiness is not keeping pace with adoption: a 2025 survey of 400 UNESCO Chairs and UNITWIN Networks across 90 countries found that while nine in ten respondents reported using AI tools in their professional work, only 19% of their institutions had a formal AI policy in place, and a further 42% reported that guiding frameworks were under development (UNESCO, 2025).

In Latin America and the Caribbean (LAC), this gap between adoption and institutional governance is particularly pronounced. Drawing on regional data covering 33 countries, the Latin America AI Index (ILIA) found that LAC accounts for 14% of global visits to AI tools despite representing only 11% of global internet users, indicating a level of end-user adoption that outpaces the region's digital weight (Soto et al., 2025). Yet no country in the region exceeds the world average in AI investment relative to GDP per capita, with the regional average remaining six times below that threshold (Soto et al., 2025). This asymmetry is compounded by weak national regulatory frameworks, fragmented institutional policy, and a structural investment deficit that makes the governance challenge particularly acute (Azevedo et al., 2025; Soto et al., 2025).

The present study contributes institution-level evidence to this regional picture, drawing on survey data from 200 higher education institutions across 19 countries in LAC. This adoption dynamic plays out differently across the core missions of the university. The three-mission model, which distinguishes teaching and learning, research, and community engagement as the three core purposes of higher education (Jongbloed et al., 2008), together with administration as a fourth analytically distinct dimension, provides the framework for mapping how AI integration is unfolding across institutions. Governance is treated as a fifth, cross-cutting dimension that shapes how AI adoption unfolds across all others, referring to the structures, rules, norms, and decision-making processes through which authority is exercised, accountability is maintained, and institutional direction is set (Bleiklie & Kogan, 2007).

This adoption dynamic must also be read alongside a growing body of concern within academic communities about the risks of AI integration. Frequent and unreflective reliance on AI tools has been associated with reduced critical thinking and long-term memory retention (Bai et al., 2023; Tian & Zhang, 2025). Generative AI has fundamentally complicated the terrain of academic integrity, with students frequently failing to conceptualise AI-assisted work as equivalent to traditional plagiarism and faculty often unable to detect it or reach consistent judgments about acceptable use (Kofinas et al., 2024; Lund et al., 2025).

AI systems trained predominantly on English-language and Western-centric content carry documented risks of algorithmic bias that are heightened in the LAC context by the region's linguistic and cultural diversity (Guo et al., 2024; Kofinas et al., 2024). Concerns about data privacy, the commercialization of educational infrastructure through dependency on private providers, and the environmental cost of large-scale AI deployment compound these challenges (Klimova & Pikhart, 2025; Lindebaum et al., 2025).

This article presents the most significant findings of a broader regional study to be published in full in 2026, based on a survey of 200 higher education institutions across 19 countries in LAC conducted between August and October 2025. It maps AI adoption across four dimensions (teaching and learning, research, community engagement, and administration) and examines governance as a fifth, cross-cutting dimension. The article is organised as follows: Section 2 describes the methodology; Section 3 presents the results, covering governance first and then patterns of adoption across the institutional missions and administration; and Section 4 discusses the cross-cutting findings and their implications.

2. METHODOLOGY  

2.1 Theoretical Framework

The study adopts an analytical framework grounded in two complementary bodies of theory. The first is the three-mission model of the contemporary university, which distinguishes teaching and learning, research, and community engagement as the three core purposes through which higher education institutions serve the public interest (Jongbloed et al., 2008). This model is reaffirmed by the UNESCO 2026 World Higher Education Conference roadmap, which describes these three purposes as interrelated and notes that universities increasingly leverage their overlap (UNESCO, 2026). The framework treats administration as a fourth, analytically distinct dimension. Not a core mission, but the organisational infrastructure through which missions are executed. AI applications in administration operate primarily on institutional data and shape the conditions under which students and staff experience the university, carrying distinct ethical and governance implications from those in teaching or research (Khairullah et al., 2025; Molina & Medina, 2025).

Governance is treated as a fifth, cross-cutting dimension that shapes how AI adoption unfolds across all others. Governance refers to the structures, rules, norms, and decision-making processes through which authority is exercised, accountability is maintained, and institutional direction is set (Bleiklie & Kogan, 2007). The distinction between administration and governance matters in the context of AI: administrative decisions about whether to deploy an early warning system may be made by management, but whether such a system is accountable, equitable, and aligned with institutional values is a governance question.

Rogers' Diffusion of Innovations theory complements this by distinguishing between initial adoption and subsequent institutionalisation (García Aviles, 2020; Jin et al., 2025; Khairullah et al., 2025), which informed the decision to assess not only whether institutions have adopted AI but whether the organisational conditions necessary for sustained and accountable integration are in place.

2.2 Research design

This study adopts a quantitative, descriptive research design. The aim is to produce a mapping of the current state of AI adoption and governance across higher education institutions in LAC, interpreted through the analytical lens of the five dimensions described above.

The study is exploratory in nature, and findings are presented as descriptive evidence of patterns across a diverse institutional sample rather than as generalisable estimates for the region as a whole. The survey was distributed through open dissemination channels without a pre-defined sample frame, and no claims of statistical representativeness are made. Data analysis is descriptive throughout, examining frequencies and distributions across two main variables: institutional type (public, private non-profit, and private for-profit) and academic orientation (multidisciplinary, polytechnic, and specialised).

2.3 Survey Instrument

Data were collected through a structured online survey comprising 33 closed questions, organised around the five analytical dimensions of the study: teaching and learning, research, community engagement, administration, and governance. Where response options included an open "other" category, responses were reviewed during data cleaning and reclassified into the existing closed categories where appropriate or excluded if they fell outside the scope of the question. The instrument was reviewed by experts at the United Nations University Institute for the Advanced Study of Sustainability (UNU-IAS) prior to administration, providing an external consultation on content validity and relevance. The survey was administered between August and October 2025 through multiple channels: direct email outreach to UNESCO IESALC institutional networks and dissemination through LinkedIn and the UNESCO IESALC newsletter.

2.4. Respondents and Data Analysis

The survey targeted senior institutional leaders responsible for governance-related functions. Eligible respondent profiles included: President, Rector, or Vice-Chancellor; Vice President or Deputy Head (Academic, Research, or Administration); Dean or Head of Faculty or Department; and Head of Institutional Planning or Strategy. The survey was not administered to teaching staff or students, as the study sought institution-level perspectives on AI adoption and governance from those with decision-making authority.

A total of 375 responses were received. These underwent a manual data cleaning process guided by the following exclusion criteria: responses from individuals outside the eligible respondent profile; incomplete responses; responses from institutions located outside the LAC region; and cases where more than one response was received from the same institution. In the latter case, only the response from the most senior or governance-relevant respondent was retained. Following this process, 200 valid responses were included in the final analysis. These 200 responses represent institutions across 19 countries in the region. Colombia is the most represented country with 40 institutions (20%), followed by México (n=30, 15%) and Argentina (n=27, 14%). Together, these three countries account for approximately half of the sample. The remaining 16 countries each contribute between 1% and 8% of responses. To contextualise the scale of this sample, the World Higher Education Database records 4,172 universities across Latin America and the Caribbean (International Association of Universities, 2026), meaning the survey captures approximately 4.8% of the total institutional population.

In terms of institution type, public institutions account for 43.5% of the sample, private non-profit institutions for 46.0%, and private for-profit institutions for the remaining 10.5% (Graph 1). Institutions were classified by academic orientation following a systematic review of each institution's official webpage. Institutions offering a broad range of programmes across multiple disciplines were classified as Multidisciplinary (n=154, 77%). Institutions with a clear technological or polytechnic orientation were classified as Polytechnic (n=16, 8%). All remaining institutions with a narrower domain-specific focus were consolidated under Specialised (n=30, 15%) (Graph 2). Findings related to polytechnic universities should be interpreted with caution, given the small subgroup size and the fact that 75.0% of polytechnic universities in the sample are public institutions, meaning that some patterns observed in this group may partly reflect their predominantly public governance rather than their academic orientation alone.

 

Graph 1. Distribution by institution type

(%)

Graph 2. Distribution by academic orientation

 (%)

Fountain: Own elaboration

3. RESULTS

Before presenting the results, two points about the scope of the findings should be noted. While the survey reached institutions across 19 countries in the region, responses are unevenly distributed: Colombia (n=40, 20%), México (n=30, 15%), and Argentina (n=27, 14%) together account for nearly half the sample, reflecting both the size of their higher education systems and the reach of the dissemination channels used. As described in the methodology, the survey was distributed through open dissemination channels without a pre-defined sample frame, and no claims of statistical representativeness are made. The findings should therefore be read as descriptive evidence of patterns across a diverse set of institutions rather than as generalisable estimates for LAC as a whole.

3.1. Institutional Readiness and Governance of AI

The governance findings are the most consequential in the survey, as they define the conditions under which all patterns of adoption documented below are taking place. The picture that emerges is one of widespread AI integration proceeding in the absence of the institutional frameworks required to govern it responsibly.

When asked to identify what enables AI adoption at their institution, university leaders pointed to people rather than resources or policy. Internal working groups and champions (71.0%), staff training initiatives (65.5%), and committed university management (52.0%) outpaced structural enablers such as digital infrastructure (26.0%), external partnerships (9.5%), or national policy incentives (9.0%). This people-driven adoption dynamic helps explain both the pace of change and its fragility. National policy regulation and financing availability are cited by fewer than 10% of respondents, suggesting that university leaders do not perceive the broader policy environment as a meaningful driver of AI adoption at the institutional level. This is consistent with Rogers' Diffusion of Innovations framework, which identifies innovators and early adopters within institutions as the primary transmission mechanism for new technologies (García Aviles, 2020; Jin et al., 2025), but also warns that adoption driven by champions alone, without reaching the stage of institutional routinisation, remains vulnerable to disruption.

3.1.1. Institutional Strategy. Only 26.0% of institutions report having a formal strategy that explicitly includes AI, while a further 32.0% have a broader digital strategy that does not specifically address AI. The largest single group, 41.0%, has no formal strategy of any kind, meaning that for a significant share of universities in the region, AI integration is proceeding without an institutional framework to guide it. This mirrors patterns documented in high-income countries, where nearly a third of top-tier universities across Australia, Canada, China, the UK, and the US similarly lacked any formal AI policy (Parker et al., 2025). A UNESCO survey of institutions hosting a UNESCO Chair or UNITWIN Network found that around 45% of LAC institutions had or were developing guidance, compared to around 70% in Europe and North America (UNESCO, 2025). The findings of the present study suggest that the institutions surveyed fall at or slightly below this LAC regional average.

The contrast between institution types is one of the most pronounced findings in the survey. Non-profit private universities are almost three times as likely as public universities to have a formal AI strategy (39.1% vs 13.8%). Public universities appear to be engaging with the AI agenda through broader digital strategies (35.6%) rather than AI-specific ones, which may also reflect concerns about equity and societal risks. Specialised universities are the most strategically underprepared, with 63.3% reporting no formal strategy of any kind.

3.1.2. Budget and Funding. Financial investment in AI is limited across the region. Overall, only 8.0% of institutions report having a specific budget dedicated to AI initiatives, while more than half (55.0%) report no specific AI budget at all. A further 29.5% finance AI activities within a broader digital transformation or innovation budget. The public-private divide is sharp: public universities are the least likely to have any form of dedicated AI funding, with 67.8% reporting no specific budget. They are by far the most reliant on national government funding (49.4%), pointing to a structural dependency that may make their AI investment more contingent on external policy cycles than on internal institutional commitment. Non-profit private universities rely overwhelmingly on central institutional budgets (92.4%), suggesting greater financial self-determination. The pattern is consistent with what the Latin America AI Index (ILIA) 2025 documents at the national level, as no country in the region exceeds the world average in AI investment relative to GDP per capita, with the regional average six times below that threshold (Soto et al., 2025).

3.1.3. AI Policies and Academic Integrity. The institutional policy landscape for AI in higher education across the region is still in formation. Only 18.5% of universities have institution-wide AI policies, and a further 21.0% have policies covering specific areas such as teaching or research. The largest single group, 29.0%, reports that policies are currently under development, while 15.0% report no policies of any kind. The non-profit private universities are more than twice as likely as public institutions to have institution-wide policies (26.1% vs 11.5%). Notably, 19.5% of public universities report that individual faculty members have developed their own informal guidelines, the highest of any group, which may reflect a bottom-up dynamic in the absence of top-down institutional frameworks. As Azevedo et al. (2025) document, many institutional AI guidelines in LAC are disjointed, hard to enforce, and quickly outdated, leaving faculty to navigate AI use independently and without cross-departmental alignment.

Among the 79 universities that do have policies, teaching and learning (79.7%) and academic integrity (79.7%) are the most addressed areas, followed by research ethics (72.2%) and data privacy and cybersecurity (41.8%). Areas such as equity and non-discrimination (26.6%) and administrative decision-making (26.6%) are less commonly addressed, suggesting that the policy conversation has so far focused on the academic rather than the broader institutional and societal dimensions of AI governance.

On academic integrity specifically, only 24.0% of universities report having guidelines on student AI use that are clear and actively communicated to the student body. The largest group, 46.5%, reports that formal guidelines do not yet exist but are under discussion, while 15.5% report no guidelines at all. The behavioural consequences of this policy gap are visible as nearly half of all universities (48.5%) report increased student confusion about what constitutes appropriate AI use, and 35.0% report increased plagiarism cases. Faculty integrity concerns are equally significant. 40.0% of universities report concerns about equity and consistency in AI-assisted assessment, 38.0% report faculty confusion about appropriate AI use, and 26.0% report increased use of AI without disclosure. These findings are consistent with the broader literature on generative AI and academic integrity, which documents that the challenge is not only one of policy absence but of communication, and that generative AI has transformed academic integrity from a plagiarism problem into an authorship problem requiring a fundamental rethinking of assessment design (Kofinas et al., 2024).

 

Graph 3. Observed Academic Integrity Consequences — Students and Faculty (% of all institutions, n=200)

Students                                                                     Faculty

Fountain: Own elaboration

Note: Respondents could select “all that apply”.

 

3.1.4. Governance Structures and Accountability. Formal AI governance remains the exception rather than the rule. Only 18.0% of universities report having a dedicated office or unit responsible for AI oversight, while the majority either distribute responsibilities across departments (43.0%) or have no designated responsibility at all (36.0%). Public universities are notably weakest on this dimension, with 48.3% reporting no designated responsibility for AI governance, compared to 25.0% of non-profit private universities.

Among the 120 universities that reported some form of governance arrangement, academic and teaching staff are the most consistently represented (81.7%), followed by IT and technical staff (66.7%) and administrative and strategic planning staff (60.8%). Most significantly, student representatives feature in only 8.3% of governance arrangements overall, and not at all in for-profit private or specialised universities.

 

Table 1. Who Sits in AI Governance Structures? (% of universities with some governance arrangement, n=120)

Actor in governance structure

Overall (n=120)

Public (n=41)

Non-profit private (n=66)

Academic / teaching staff

81.7%

85.4%

86.4%

IT / technical staff

66.7%

61.0%

69.7%

Administrative/strategic planning

60.8%

63.4%

60.6%

External consultants or partners

15.0%

14.6%

16.7%

Student representatives

8.3%

17.1%

4.5%

Fountain: Own elaboration

 

The UNESCO Recommendation on the Ethics of AI articulates inclusive, multi-stakeholder governance of AI systems as a foundational principle, requiring the inclusion of all stakeholder groups impacted by these technologies (UNESCO, 2022). The active engagement of students alongside faculty and administrators is also stressed by the OECD, which notes that the compartmentalisation of governance decisions among a narrow set of actors is an obstacle to responsible innovation (OECD, 2023).

3.1.5. Monitoring and Evaluation. The capacity to monitor and evaluate AI implementation is paramount for enabling institutions to understand how AI is affecting learning and operational outcomes and to make informed decisions. The survey findings reveal how underdeveloped this function remains. Overall, only 9.0% of universities report having formal evaluation mechanisms in place for AI implementation, while 26.5% rely on informal or project-based evaluation, and the largest group, 56.5%, reports no monitoring or evaluation at all.

Situating these figures against the adoption picture makes the gap concrete. 86.5% of universities report using AI in at least one functional area, yet only 9.0% have formal evaluation mechanisms in place, meaning the region has roughly ten times more universities deploying AI than universities systematically measuring its outcomes. Among the 147 universities using AI in teaching and learning, the area of highest adoption and most direct student impact, 49.0% have no monitoring in place. By institution type, the monitoring gap is most pronounced among public universities, of which 62.1% report no monitoring at all.

3.2. AI Adoption Across the Institutional Missions and Administration

AI adoption across the three core missions and the administrative function follows a clear hierarchy that tracks closely with the proximity of each dimension to the tools students and faculty encounter in their daily work. Teaching and learning lead, followed by research, administration, and community engagement.

3.2.1. Teaching and Learning. Teaching and learning is the mission where AI adoption is most widespread, with 73.5% of universities reporting current use. Adoption rates vary across institution types: private non-profit universities show the highest levels at 83.7%, compared to 67.8% for public institutions and 52.4% for private for-profit institutions. Across academic orientations, adoption rates are remarkably similar: multidisciplinary (74.0%), polytechnic (75.0%), and specialised (70.0%) universities all cluster within five percentage points of each other, one of the few areas in the survey where academic orientation produces no meaningful difference, suggesting that the penetration of AI in teaching and learning has been sufficiently broad to reach all institutions regardless of their academic profile.

The technology landscape is dominated by generative AI: among universities using AI in teaching, 89.8% report using tools such as ChatGPT, Copilot, or Gemini, a rate that is uniform across all institution types. This uniformity suggests that generative AI has become the default entry point for AI adoption in teaching and learning across the region, likely driven by the free or low-cost availability of these tools and their ease of use without requiring significant institutional infrastructure. Beyond generative AI, differences emerge by academic orientation: chatbots and virtual assistants are widely used by multidisciplinary (53.5%) and polytechnic universities (58.3%), but considerably rarer among specialised institutions (14.3%). Polytechnic universities stand out for their comparatively high use of adaptive learning platforms (41.7%) and AI assistive technology for students with disabilities (33.3%).

Promoting pedagogical innovation and digital transformation is the most frequently cited objective (71.4%), but this sits alongside concrete educational goals: improving teaching quality (59.9%), improving learning outcomes (54.4%), increasing student engagement (51.0%), and reducing teacher workload (50.3%).

3.2.2. Research. AI adoption in research stands at 57.0%, but the analytically significant finding is the visibility gap: 22.5% of senior institutional leaders responded that they did not know whether AI was being used in research at their own institution. This could imply that AI in research is diffusing at the level of individual researchers and groups without the strategic guidance of leadership. The distribution of "don't know" responses is broadly uniform across institution types (public 24.1%, non-profit private 21.7%, for-profit private 19.0%), suggesting this is a systemic feature of how AI is spreading in research rather than a problem concentrated in any institutional category.

Among the 114 universities that report AI use in research, the objectives are practical and efficiency-oriented: accelerating data processing and analysis (69.3%), improving research quality (64.9%), and improving efficiency in literature review and knowledge synthesis (64.0%) dominate. These findings are consistent with global evidence on how generative AI is transforming research (Molina & Medina, 2025) and with the growing literature on AI applications for literature synthesis, data analysis, and writing support in contexts where research support infrastructure remains uneven (Molina & Medina, 2025).

3.2.3. Community Engagement. Community engagement records both the lowest adoption rate of any mission at 20.0% and the highest "don't know" rate at 31.5%, reflecting its position as the dimension most distant from daily academic work and least visible to institutional leadership. By institution type, non-profit private universities show the highest adoption (27.2%), while public universities show the lowest adoption (13.8%) and the highest "don't know" rate (41.4%), which may reflect the more decentralised nature of community engagement activities in larger public universities, where initiatives may be driven by individual faculties rather than by institutional strategy.

3.2.4. Administration. AI adoption in administrative functions is reported by just over a third of surveyed institutions (34.1%). It is accompanied by a 28.5% "don't know" rate, a figure that rises to 37.9% among public universities. When more than one in four senior leaders cannot confirm whether AI is being used in their own institution's administrative operations, this could point to a dynamic in which AI tools are adopted at the departmental or operational level outside the visibility and accountability of central leadership, a pattern consistent with the governance deficit documented throughout this report.

By academic orientation, specialised universities report the highest administrative AI adoption (50.0%), above multidisciplinary (32.5%) and polytechnic universities (18.8%). One plausible explanation is that smaller institutions may face lower barriers to deploying AI tools in specific administrative functions than larger institutions with more complex systems. Among the 68 universities reporting administrative AI use, applications are concentrated in student-facing functions: student support and advising (51.5%) and admissions and enrolment (42.6%) are the two most common areas. The objectives are dominated by efficiency considerations: automation of routine tasks is cited by 79.4% of users, the most frequently cited objective, consistent with observations across the literature on administrative AI adoption in higher education (Khairullah et al., 2025; Molina & Medina, 2025).

 

Graph 4 - AI Adoption by analysed dimension (n=200)

Fountain: Own elaboration

3.3. Training Opportunities

A large majority of universities (82.5%) report having implemented AI training programmes for academic staff, indicating that capacity development for teaching personnel is a relatively common practice across the region. However, coverage drops significantly when it comes to administrative and technical staff, with only 41.0% extending training to this group. This gap is particularly concerning as AI adoption in administrative functions is already occurring, and administrative staff who encounter AI tools in their daily work without appropriate training may represent a governance risk.

AI training for students is less widespread than for staff, and the modalities through which it is delivered vary. The most common approach is the elective course model (41.0%), followed by curriculum integration (33.5%), while 34.5% of universities report no student-facing AI training at all. The prevalence of the elective model over full curriculum integration suggests that many universities are at an early stage, offering AI courses as an optional addition rather than as a core competency embedded across programmes (Jaramillo & Chiappe, 2024). By institution type, public universities show the highest share reporting no training (40.2%). The most interesting finding by academic orientation is the contrast between polytechnic universities, which report the highest rate of curriculum integration (56.2%), and specialised universities, which show the lowest rates of any training modality and the highest share reporting no student AI training (56.7%).

Among the 147 universities that report using AI in teaching and learning, 8.2% have no academic staff training programme in place, and 24.5% offer no student training of any kind, meaning that a significant share of institutions are deploying AI in the classroom without having prepared either those who teach with it or those who learn through it. This training gap resonates with global findings: the Digital Education Council Faculty Survey found that 78% of faculty across 28 countries do not consider their institutions to have provided sufficient resources for developing AI literacy (Digital Education Council, 2025).

3.4. Challenges to AI Integration

Lack of financial resources emerged as the top-ranked challenge by a wide margin, cited as the most important obstacle by 37.5% of respondents, more than double the rate of any other challenge. Insufficient staff expertise and limited technical infrastructure constitute the second priority group and should be read together rather than in strict sequence. Notably, the lack of clear policy guidelines ranks sixth overall despite the significant policy gap documented in earlier sections. This apparent paradox may suggest that institutions are proceeding with AI adoption regardless of the policy environment, consistent with the broader literature documenting a pattern in which institutional AI adoption consistently outpaces governance frameworks (Jin et al., 2025; Pireci Sejdiu & Sejdiu, 2025; UNESCO IESALC, 2023).

 

Table 2. Ranking of Challenges to AI Integration (n=200)

Challenge

Average Rank (1=most important)

Lack of financial resources

3.26

Insufficient staff expertise

3.62

Limited technical infrastructure

4.01

Ethical/governance concerns

4.07

Resistance to change

4.54

Lack of policy guidelines

4.71

Data privacy/cybersecurity risks

5.78

Uncertainty about long-term impact

6.01

Fountain: Own elaboration.

Note: Respondents ranked all eight challenges from 1 (most important) to 8 (least important). Lower average rank indicates a more commonly cited obstacle.

 

4. DISCUSSION

4.1. Cross-cutting Findings

The findings from this study reveal a regional higher education landscape in which AI adoption is accelerating while the institutional conditions required to govern it responsibly remain significantly underdeveloped. Four cross-cutting findings stand out as analytically consequential.

First, the adoption hierarchy reflects structural drivers rather than deliberate institutional prioritisation. Teaching and learning lead at 73.5%, followed by research (57.0%), administration (34.1%), and community engagement (20.0%). This hierarchy tracks closely with the proximity of each dimension to the tools students and faculty use daily, particularly generative AI, reported by 89.8% of universities using AI in teaching. AI has entered universities primarily through everyday academic work, driven by the wide availability of low-cost tools rather than by institutional planning. Community engagement, the mission most distant from daily academic work, is also the one most invisible to senior leadership, with the highest "don't know" rate in the survey. This pattern is consistent with what UNESCO IESALC has observed directly across more than 40 higher education institutions in the region: AI adoption in teaching and learning tends to follow a bottom-up dynamic, where students begin using AI tools, faculty subsequently adapt their practices, and institutional governance responds by developing guidelines (UNESCO IESALC, 2025a).

Second, the public-private divide is the most consistent and consequential finding in the survey. Non-profit private universities outperform public institutions on every single dimension measured: adoption rates across all four functional areas, formal strategy, dedicated budget, institution-wide policies, academic integrity guidelines, governance structures, staff training, and student training. The gap is most pronounced in formal AI strategy (39.1% vs 13.8%) and the share of public universities reporting no student training at all (40.2% of public vs 8.6% of non-profit private universities). This finding carries equity implications that extend beyond higher education itself. Public universities in the region carry the largest student populations, the broadest social mandates, and serve students from lower-income backgrounds who stand to benefit most from the potential of AI tools to provide personalised support and early intervention (Wong et al., 2025). A persistent and widening AI readiness gap between public and private institutions therefore risks compounding existing structural inequalities in the higher education system.

 

Graph 5 - The Public–Private Divide Across Six Key Dimensions (% of institutions by type)

Fountain: Own elaboration

Note: For-profit private (n=21) excluded given small n.

 

Third, specialised universities emerge as the most consistently vulnerable group across the survey. Across every governance dimension, formal strategy (63.3% with no strategy), dedicated budget (76.7% with no specific AI budget), institution-wide policies (only 10.0%), staff AI training (70.0%), and student training (56.7% with no training at all), specialised universities come out weakest. This is particularly concerning because these are institutions that train students for specific professional fields, including health sciences, teacher education, law, and business, where the ethical and practical implications of AI for future professional practice are immediate and concrete. The governance deficit in specialised universities represents a potential systemic risk for the professional sectors their graduates will enter.

Fourth, students are the least protected actors in the current AI landscape. They are the primary users of AI tools, with 86% already using them globally in their studies (Digital Education Council, 2024), yet they are simultaneously the group least likely to receive formal training, least represented in governance structures, and most directly affected by the academic integrity consequences of inadequate institutional guidance. Students appear in governance structures in only 8.3% of institutions overall. The UNESCO Recommendation on the Ethics of AI makes clear that inclusive multi-stakeholder governance is not optional but foundational to responsible AI integration (UNESCO, 2022). The current governance model, concentrated predominantly among academic and IT staff, fails to meet this standard.

4.2. The Adoption-Governance Gap and its Structural Drivers

The central finding of this study is the gap between the scale of AI adoption and the maturity of the institutional frameworks governing it. 86.5% of universities report using AI in at least one functional area, yet only 26.0% have a formal AI strategy, 18.5% have institution-wide policies, 8.0% have a dedicated AI budget, 18.0% have a designated governance structure, and just 9.0% have formal evaluation mechanisms. Expressed differently, the region has roughly ten times more universities deploying AI than universities systematically measuring what that deployment is doing.

The mismatch between what drives adoption and what would sustain it responsibly is a structural vulnerability. University leaders point to internal working groups and champions (71.0%), staff training initiatives (65.5%), and committed management (52.0%) as the main enablers of AI adoption. National policy incentives and external financing feature in fewer than 10% of responses, and financing availability is cited by only 3.0% as an enabling factor, despite financial constraints being ranked as the single most important obstacle to AI integration. Taken together, this evidence supports the characterisation that AI adoption in the region is being driven by motivated individuals operating within institutions that have not yet built the frameworks to sustain, govern, or evaluate what those individuals are achieving.

The monitoring deficit makes accountability difficult in ways that extend beyond institutional management. Universities that do not systematically evaluate their AI implementation cannot demonstrate to students, staff, or external stakeholders whether AI is achieving its intended purposes, cannot identify whether AI tools are widening or narrowing equity gaps, and cannot contribute to the regional evidence base that policymakers need to make informed decisions. The 56.5% rate of no monitoring at all means that the region is scaling AI implementation without a systematic way of understanding what that scaling is producing. This is a structural accountability problem.

The regional structural context amplifies these institutional gaps. As the ILIA 2025 documents, LAC countries are characterised by high adoption and low investment: the region represents 6.6% of global GDP but attracts only 1.12% of global AI investment, and most national AI strategies lack the financing, implementation mechanisms, and evaluation systems that would give them operational force (Soto et al., 2025). In the absence of national frameworks providing guidance, standards, and accountability mechanisms, the institution becomes the primary site of governance for AI adoption. Where institutional governance capacity is also underdeveloped, as documented throughout this study, the result is a situation in which the entire weight of responsible AI integration falls on individual champions whose knowledge and commitment cannot substitute for the structural conditions they are operating without.

5. CONCLUSIONS

This study maps AI adoption and governance across 200 higher education institutions in LAC, providing an institution-level picture of where AI is being used, how it is being governed, what capacities exist, and what challenges remain. While the survey was extended to different institutions in the region, most of the responses came from Colombia (n=40, 20%), México (n=30, 15%), and Argentina (n=27, 14%). As stated in the methodology, the study employed a census-style open dissemination strategy rather than a probabilistic sampling approach, and no claims of statistical representativeness are made. The findings should therefore be read as descriptive evidence of patterns across a diverse set of institutions rather than as generalisable estimates for LAC. With this scope in mind, the results document a regional higher education system in which AI adoption has already accelerated, driven by student and faculty demand and the commercial availability of tools, while the governance infrastructure required to make that adoption responsible, equitable, and accountable remains underdeveloped at both national and institutional levels.

The cross-cutting findings point to four priority areas for institutional action: closing the adoption-governance gap through formal strategy, dedicated governance structures, and monitoring mechanisms; addressing the public-private divide through targeted investment and policy frameworks that support public universities in building AI readiness proportionate to their social mandates; protecting students through AI literacy integrated into curricula, clear and communicated academic integrity guidelines, and formal student representation in governance; and building the regional evidence base through shared monitoring frameworks and periodic reviews that make the outcomes of AI integration visible at both institutional and policy levels. UNESCO's higher education roadmap calls for higher education institutions to adopt governance structures capable of ensuring that digital technologies serve equity rather than undermine it (UNESCO, 2026). In LAC, the distance between that aspiration and current institutional reality is the central finding of this study.

The patterns documented in this study must be situated within a broader critical conversation about what AI integration in higher education entails. The risks associated with AI adoption are well evidenced: cognitive dependency and reduced critical thinking where AI tools are introduced into learning environments (Bai et al., 2023; Tian & Zhang, 2025), the transformation of academic integrity from a plagiarism problem into an authorship problem that existing institutional frameworks were not designed to address (Kofinas et al., 2024), the replication of algorithmic bias in student-facing systems trained on data that reflects structural inequalities (Gándara et al., 2024), and the gradual ceding of epistemic governance to private technology companies whose commercial incentives do not necessarily align with public educational values (Lindebaum et al., 2025). In the LAC context, where most AI tools are trained predominantly on English-language and Western-centric content, the risk of epistemic and cultural homogenisation carries weight for universities with explicit mandates to engage with and serve diverse, historically marginalised communities. These concerns deserve serious and ongoing institutional engagement.

Several limitations of this study should be acknowledged. Several limitations of this study should be acknowledged. Data were collected through open dissemination channels without a pre-defined sample frame, and participation was shaped by the reach of the channels used, which means findings cannot be considered statistically representative of higher education institutions across LAC. Moreover, the rapidly evolving nature of AI means that the landscape mapped here reflects a specific moment in time (August to October 2025), and some findings may already have shifted.

These limitations point toward productive directions for future research. The descriptive picture produced by this study provides a foundation that qualitative methods are well positioned to deepen in-depth interviews with institutional leaders, faculty, and students would allow the patterns documented here to be examined from multiple institutional perspectives and would help explain the mechanisms behind the adoption-governance gap. Comparative studies with other world regions would situate the LAC findings within a global picture and help identify whether the structural vulnerabilities documented here are region-specific or part of a broader global pattern.

6. REFERENCES

Azevedo, L., Robles, P., Best, E., & Mallinson, D. J. (2025). Institutional policies on artificial intelligence in higher education: Frameworks and best practices for faculty. New Directions for Adult and Continuing Education, 2025(188), 70–78. https://doi.org/10.1002/ace.70013

Bai, L., Liu, X., & Su, J. (2023). ChatGPT: The cognitive effects on learning and memory. Brain-X, 1(3), Article e30. https://doi.org/10.1002/brx2.30b

Baker, J. (2012). The technology-organization-environment framework. In Y. K. Dwivedi, M. R. Wade, & S. L. Schneberger (Eds.), Information systems theory (Vol. 28, pp. 231–245). Springer. https://doi.org/10.1007/978-1-4419-6108-2_12

Bleiklie, I., & Kogan, M. (2007). Organization and governance of universities. Higher Education Policy, 20(4), 477–493. https://doi.org/10.1057/palgrave.hep.8300167

Digital Education Council. (2024). Digital Education Council global AI student survey 2024. https://www.digitaleducationcouncil.com/post/digital-education-council-global-ai-student-survey-2024

Digital Education Council. (2025). Digital Education Council global AI faculty survey 2025. https://www.digitaleducationcouncil.com/post/digital-education-council-global-ai-faculty-survey

Gándara, D., Anahideh, H., Ison, M. P., & Picchiarini, L. (2024). Inside the black box: Detecting and mitigating algorithmic bias across racialized groups in college student-success prediction. AERA Open, 10. https://doi.org/10.1177/23328584241258741

García Avilés, J. A. (2020). Diffusion of innovation. In K. Matthes (Ed.), The International Encyclopedia of Media Psychology. Wiley. https://doi.org/10.1002/9781119011071.iemp0137

Guo, H., Narayanan Venkit, P., Jang, E., Srinath, M., Zhang, W., Mingole, B., Gupta, V., Varshney, K. R., Sundar, S. S., & Yadav, A. (2024). Hey GPT, can you be more racist? Analysis from crowdsourced attempts to elicit biased content from generative AI. arXiv. https://doi.org/10.48550/arXiv.2410.15467

International Association of Universities. (2026). World Higher Education Database (WHED). IAU; UNESCO. https://whed.net/home.php

Jaramillo, J. J., & Chiappe, A. (2024). The AI-driven classroom: A review of 21st-century curriculum trends. Prospects, 54(3), 645–660. https://doi.org/10.1007/s11125-024-09704-w

Jin, Y., Yan, L., Echeverria, V., Gašević, D., & Martinez-Maldonado, R. (2025). Generative AI in higher education: A global perspective of institutional adoption policies and guidelines. Computers and Education: Artificial Intelligence, 8, 100348. https://doi.org/10.1016/j.caeai.2024.100348

Jongbloed, B., Enders, J., & Salerno, C. (2008). Higher education and its communities: Interconnections, interdependencies and a research agenda. Higher Education, 56(3), 303–324. https://doi.org/10.1007/s10734-008-9128-2

Khairullah, S. A., Harris, S., Hadi, H. J., Sandhu, R. A., Ahmad, N., & Alshara, M. A. (2025). Implementing artificial intelligence in academic and administrative processes through responsible strategic leadership in higher education institutions. Frontiers in Education, 10, 1548104. https://doi.org/10.3389/feduc.2025.1548104

Klimova, B., & Pikhart, M. (2025). Exploring the effects of artificial intelligence on student and academic well-being in higher education: A mini-review. Frontiers in Psychology, 16, 1498132. https://doi.org/10.3389/fpsyg.2025.1498132

Kofinas, A., Han Hue Tsai, C., & Pike, D. (2024). The impact of generative AI on academic integrity of authentic assessments within a higher education context. British Journal of Educational Technology, 56(6), 2522–2549. https://doi.org/10.1111/bjet.13585

Lindebaum, D., Nolan, E., Ashraf, M., Islam, G., & Ramirez, M. F. (2025). The transformation of epistemic agency and governance in higher education through large language models: Toward a future of organized immaturity. Organization Studies. https://doi.org/10.1177/01708406251392002

Lund, B. D., Lee, T. H., Mannuru, N. R., & Arutla, N. (2025). AI and academic integrity: Exploring student perceptions and implications for higher education. Journal of Academic Ethics, 23(3), 1545–1565. https://doi.org/10.1007/s10805-025-09613-3

Molina, E., & Medina, E. (2025). AI revolution in higher education (Digital Innovations in Education Brief No. 4). World Bank. https://doi.org/10.1596/43298

Organisation for Economic Co-operation and Development. (2023). OECD Digital Education Outlook 2023. OECD Publishing. https://doi.org/10.1787/c74f03de-en

Parker, L., Loper, A. J., Hayes, J., Karakas, A., White, S., & Hallman, H. (2025). Comparative analysis of artificial intelligence policies in universities across five countries. Discover Computing, 28(1), 267. https://doi.org/10.1007/s10791-025-09745-5

Pireci Sejdiu, N., & Sejdiu, S. (2025). The quiet transformation of higher education in the AI era. Open Research Europe, 5, 249. https://doi.org/10.12688/openreseurope.20715.1

Soto, A., Duran, R., Moreno, A., Adasme, S., Rovira, S., Jordán, V., & Poveda, L. (2025). Índice Latinoamericano de Inteligencia Artificial (ILIA) 2025 (Documentos de Proyectos LC/TS.2025/68/Rev.1). CEPAL. https://indicelatam.cl/wp-content/uploads/2025/10/Documento_ILIA_2025.pdf

Tian, J., & Zhang, R. (2025). Learners’ AI dependence and critical thinking: The psychological mechanism of fatigue and the social buffering role of AI literacy. Acta Psychologica, 260, 105725. https://doi.org/10.1016/j.actpsy.2025.105725

United Nations Educational, Scientific and Cultural Organization. (2022). Recommendation on the ethics of artificial intelligence. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000381137

United Nations Educational, Scientific and Cultural Organization. (2025). UNESCO survey: Two-thirds of higher education institutions have or are developing guidance on AI use. https://www.unesco.org/en/articles/unesco-survey-two-thirds-higher-education-institutions-have-or-are-developing-guidance-ai-use

United Nations Educational, Scientific and Cultural Organization. (2026). Transforming higher education: Global collaboration on visioning and action. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000397582

United Nations Educational, Scientific and Cultural Organization, International Institute for Higher Education in Latin America and the Caribbean. (2023). Harnessing the era of artificial intelligence in higher education: A primer for higher education stakeholders. UNESCO-IESALC. https://unesdoc.unesco.org/ark:/48223/pf0000386670

United Nations Educational, Scientific and Cultural Organization, International Institute for Higher Education in Latin America and the Caribbean. (2025a). The challenges of AI in higher education and institutional responses: Is there room for competency frameworks? UNESCO-IESALC. https://unesdoc.unesco.org/ark:/48223/pf0000394935

United Nations Educational, Scientific and Cultural Organization, International Institute for Higher Education in Latin America and the Caribbean. (2025b). The role of higher education in national artificial intelligence strategies: A comparative policy review. UNESCO-IESALC. https://unesdoc.unesco.org/ark:/48223/pf0000392047

Wong, W., Aristidou, A., & Scheuermann, K. (2025). The future of learning or the future of dividing? Exploring the impact of generative artificial intelligence on higher education. Data & Policy, 7. https://doi.org/10.1017/dap.2025.10011

 

Cómo citar (APA):

Valentini, A. (2025). AI Adoption and Governance in Latin American and Caribbean Higher Education: Findings from a Regional Survey. Revista Educación Superior y Sociedad (ESS), 37(2), 288-307. DOI:10.54674/ess.v37i2.1278