By Javier Surasky
.png)
Discussions
of artificial intelligence in higher education are often organized around
principles such as autonomy, transparency, equality, privacy, accountability,
and human oversight. Necessary as these principles are, they leave one critical
issue largely unexplored: the institutional conditions required to put them
into practice, a part of AI digital governance.
UNESCO has
warned that “the effective and ethical use of AI in education depends on
various factors including, but not limited to, access to digital
infrastructure, and to the internet in particular; availability of AI
resources; regulations on data security and privacy; policy guidance and
incentives; and professional development opportunities.” (Miao & Cukurova,
2024, p. 22).
Putting
ethical principles into practice requires material, legal, organizational, and
professional capacities that universities do not always possess. Transparency
depends on having sufficient information about the systems being used. Privacy
requires an understanding of what data are generated, how they are processed,
and where they are stored. Human oversight presupposes that those responsible
for exercising it have adequate knowledge, time, and authority. Equality
requires accessible services, while accountability depends on identifiable
authorities, oversight procedures, and accessible and secure complaint
mechanisms.
Ethical AI
governance in higher education, therefore, faces a gap between demanding
normative principles and institutional capacities that are insufficient or
unevenly distributed.
Principles in Context
An ethical
framework for AI in higher education should establish which uses in research,
teaching, and administration are acceptable. It should also specify the
conditions, procedures, and oversight mechanisms under which those uses may
take place. As Reimers et al. (2026, p. 157) observe, “The promise of AI in
education cannot be understood in isolation, particularly in fragile systems
where initial conditions, such as infrastructure, teacher capacity, and
inequality, exert powerful influences”.
Every
educational technology enters a preexisting structure of resources, skills,
hierarchies, and inequalities that shapes how it is adopted and used. In the
case of AI, outcomes are conditioned by differences in access, understanding,
and the ability to respond when rules are breached or systems fail. A
university with technical teams, legal counsel, its own infrastructure, and the
ability to negotiate with vendors can integrate new technologies under very
different conditions from one that relies on free versions, external services,
and fragmented or poorly coordinated decision-making structures.
Applying
the same principle uniformly to institutions with different capacities can
produce unequal outcomes. Such differences arise not only between universities
but also among faculties, campuses, and administrative units within the same
institution.
Major AI
systems are developed, maintained, and modified by actors outside universities.
Their internal mechanisms are not necessarily transparent, and their terms of
use, functions, and costs may change as a result of decisions over which
educational institutions have no control.
This
problem has been widely examined in other areas of technology governance, yet
it remains insufficiently incorporated into university debates on AI.
Universities are attempting to govern technologies over which they exercise
only limited control. Several of those limitations deserve closer attention.
Infrastructure
The
infrastructure required to work with AI includes Internet access and computers,
but also processing capacity, secure storage, cloud services, stable networks,
maintenance, specialized technical support, and cybersecurity safeguards.
These
resources cannot be assessed solely in quantitative terms. Computing, storage,
and connectivity needs vary across disciplines and intended uses. Indicators
that appear equivalent may therefore conceal substantial differences in quality
and adequacy, both between universities and among faculties within the same
institution.
From this
perspective, the recommendation to prioritize generative artificial
intelligence tools “that are open access or low cost” (Sánchez Mendiola &
Carbajal Degante, 2023, p. 80) may reduce barriers to entry, but it does not
eliminate inequalities arising from differences between versions, computing
capacity, technical support, security, and access to advanced features. Nor
does it prevent new forms of stratification among institutions, faculty, and
students.
Debates on
ethics in higher education should therefore incorporate a dimension of
strategic autonomy. This should not be understood as technological
self-sufficiency, but as an institution’s capacity to evaluate systems, compare
alternatives, negotiate terms, restrict uses, choose among providers, and
replace technologies when they no longer meet the technical, legal, or ethical
requirements established by the university.
Human Capacity as Infrastructure
Capacity
building for AI should reach everyone who participates in university life:
institutional authorities, administrative teams, faculty, professional and
support staff, researchers, and students.
Training is
part of the institutional infrastructure required to govern technology. It
should not be limited to learning how to write effective prompts or use the
features available in a particular application.
Those
involved need at least a basic understanding of how AI systems work, what
biases are, how they may arise, what privacy risks these systems create, and
how they may affect teaching, assessment, research, and administration. They
also need access to institutional channels for consultation, guidance, and
training.
Training
requirements should increase in proportion to a person’s institutional
responsibilities and to the potential impact of the systems with which they
work. Using a tool to organize study materials does not require the same
capabilities as deciding whether to incorporate it into assessment, resource
allocation, or the management of students’ academic trajectories.
Training
must also be continuous and context-specific. AI systems change rapidly, add
new features, and alter their terms of use. The needs of different disciplines
and university activities also vary. Studying, teaching, conducting research,
and managing institutions with AI require different skills.
Universities
need to educate critical users, not merely certify instrumental competencies.
Transparency
In
discussions of ethics in higher education, transparency is usually framed as an
obligation imposed on students, researchers, and faculty. They may be required
to disclose that they used an AI tool, explain the purpose for which it was
used, or retain a record of the instructions given to the system. These
measures are useful, but they address only part of the problem.
Transparency
requirements should also apply to institutions and technology providers. A
university seeking to make its ethical principles operational needs to know, at
a minimum, what functions each system performs, what data it receives, how long
it retains them, whether it reuses them, where it processes them, and what
mechanisms it provides for deleting or retrieving information.
There is
little value in imposing strict rules on users if the institution then accepts
lengthy and difficult-to-understand contracts containing data reuse provisions
that it does not fully understand.
Complete
technical transparency is not possible in the field of AI. Even so, companies
seeking to provide services to universities should supply the information
requested by the institution. That information should then be assessed against
the university’s ethical principles when contracts, purchases, or
authorizations for use are being considered.
Institutions
with greater capacity can review contracts, test systems, and negotiate
specific clauses. Those without such resources are often compelled to accept
standard terms.
Interuniversity
cooperation may therefore be necessary to pool technical and legal expertise
and reduce disparities in bargaining power.
Privacy
The
educational use of AI involves continuous information flows. Students,
faculty, researchers, and administrative teams may enter academic work,
assessments, research data, personal records, and institutional information
into AI systems.
A general
warning not to upload sensitive information is insufficient, partly because
there is not always a shared understanding of what should count as sensitive
information.
Privacy
thus reappears as a question of institutional capacity. It exposes the tension
between the need to record interactions or preserve evidence and the potential
expansion of surveillance over students and university employees that such
records may enable.
This
tension is intensified by the asymmetry inherent in educational relationships.
When a platform is mandatory for taking a course, teaching, or being assessed,
accepting its terms is not a genuinely free choice. The same applies when an
institution requires its staff to use particular digital environments to
perform their work.
Human Oversight
Chan and
Colloton argue that “Adopting AI technologies into academic settings requires a
structured approach to monitoring and evaluating AI implementation” (2024, p.
150).
Human
oversight, or keeping a human in the loop, is often presented as a way to
reinforce other principles and prevent decisions from being made entirely by
machines. Under this model, a system may provide suggestions or assistance,
while final responsibility remains with a human decision-maker.
That
formula is insufficient unless institutions specify who is responsible for
oversight, what must be reviewed, and what authority the reviewer has when a
problem is identified.
Human
responsibility also requires competence, time, information, and the authority
to modify, challenge, or reject a system’s output. No one exercises meaningful
control when they are limited to approving results they do not understand, or
when the volume of decisions makes substantive review materially impossible.
It is also
necessary to distinguish among levels of risk. Using AI to organize a
bibliography is not equivalent to using it to decide whether a student may
enroll in a course, allocate resources, or assess academic performance. The
greater the potential impact on rights, academic trajectories, or resource
distribution, the stronger the requirements for documentation, oversight,
review, and training should be.
To prevent
human oversight from becoming a passive review of results, it must be
accompanied by clear and workable channels for reporting incidents and
challenging decisions. Students, faculty, researchers, and university staff
must know where to file a complaint, what procedure applies, and what remedies
or corrective measures are available.
Efforts by Argentine Universities
Argentine
universities have begun to develop internal instruments to guide the use of AI,
although their scope and level of operational detail vary.
The
National University of Cuyo adopted its Principles for the Responsible Use
of Generative AI through University Council Resolution 262/2026. The
document addresses human-centeredness, academic integrity, data protection,
accessibility, technological sovereignty, traceability, and institutional
evaluation. It also provides for support mechanisms for students, faculty, and
staff, as well as auditing criteria.
The
National Technological University adopted a more narrowly focused instrument. Resolution 279/2026 establishes guidelines for the use of generative AI in postgraduate courses and thesis writing. It includes requirements for human
oversight and control, differentiated responsibilities for faculty, students,
thesis supervisors, and evaluators, and a review of the contractual terms
governing the tools in use in order to prevent risks to intellectual property.
The
National University of the Northeast followed a different approach. Resolution 9323/2025 approved regulations governing the implementation and use of its
institutional “IA UNNE” system. The instrument links principles of
transparency, explainability, data protection, bias mitigation, and
accountability to mechanisms for informed consent, human validation,
traceability, auditing, and system updates.
At larger
and more decentralized universities, such as the University of Buenos Aires and
the National University of La Plata, responses are distributed between central
authorities and individual faculties.
At the
University of Buenos Aires, the Faculty of Economics adopted its own rules
governing the academic use of AI, while the Faculty of Law developed a pilot
program focused on AI literacy and the strategic and responsible use of
generative AI in legal practice. There is, however, no single university-wide
policy covering all faculties.
At the
National University of La Plata, an Artificial Intelligence Working Group was
established with representatives from the central administration, faculty,
researchers, and professional and support staff from different academic units.
Individual faculties have also adopted their own documents, including a
practical guide for faculty developed by the Faculty of Law and Social
Sciences. The university likewise has no general regulation applicable to all
functions and academic units.
These cases
reveal a range of responses: guiding principles, mandatory guidelines,
regulations for institutional systems, coordinating bodies, and sector-specific
recommendations. They represent meaningful progress, but the capacity to
implement them effectively, monitor compliance, and review them periodically
remains an open question.
Basic First Steps
In light of
these developments and limitations, universities need to move from general
declarations to concrete mechanisms for gathering information, making
decisions, and exercising oversight.
The first
step should be to map the systems currently in use and the functions they
perform, the data they process and the contractual terms that apply, the
existing risks and safeguards, and the distribution of capacities and
responsibilities across the institution. Without this baseline information,
ethical principles risk remaining disconnected from actual practice.
Based on
that assessment, each university can review its existing policies, address
previously unanticipated uses, identify gaps, and set priorities according to
factors such as the potential for harm, the scale of a system’s use, and its
possible effects on rights and academic trajectories.
The rapid
pace of technological change requires flexible institutional structures capable
of responding effectively. University policies should therefore provide for
periodic evaluation involving all affected groups, accessible complaint
mechanisms, and the authority to suspend uses when sufficient safeguards are
not in place.
This work
also requires coordination across faculties, especially at large and
decentralized universities. Bodies such as the Artificial Intelligence Working
Group created by the National University of La Plata, which brings together
representatives from across the university, can perform this function if they
are given clear responsibilities, adequate resources, and an operational
mandate.
Conclusion
Ethical AI
governance in higher education depends on the capacity to translate declaratory
principles into decisions, procedures, and oversight mechanisms. Doing so
requires information, infrastructure, human capacity, technical and legal
expertise, clearly assigned responsibilities, and accessible channels for
complaints and review.
It also
requires distinctions among the different areas of university activity.
Administration, research, and teaching involve different risks, actors, and
oversight needs. A general policy may establish a shared set of core
principles, but its implementation must account for these differences.
Developing
such capacities requires staff time, dedicated teams, training programs, and
sustained investment. This point must be stated clearly: without adequate
funding, universities will find it difficult to give AI governance the priority
it requires.
Without
resources, clearly assigned responsibilities, professional capacity, and
effective oversight procedures, ethical principles will remain institutional
declarations rather than functioning as meaningful standards for governing AI.
Educating
professionals and citizens for a world that is already digital requires
universities to address this issue seriously and urgently. Only then can they
fulfill their responsibility to prepare graduates to respond to the needs of
the societies in which they live and work.
References
Chan, C. K.
Y., & Colloton, T. (2024). Generative AI in higher education: The
ChatGPT effect. Routledge. https://doi.org/10.4324/9781003459026
Miao, F.,
& Cukurova, M. (2024). AI competency framework for teachers. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000391104
Reimers,
F., Azim, Z., Palomo, M.R., & Thony, C. (2026). Artificial intelligence and education in the
Global South: A systems perspective. Springer Nature Switzerland. https://doi.org/10.1007/978-3-032-11449-5