The Missing Half of AI Ethics: Universities Focus on Misconduct While Inclusion Falls Behind

Universities are rapidly establishing rules to govern artificial intelligence, but the emerging regulatory landscape reveals a troubling imbalance. Academic dishonesty, plagiarism and data protection receive extensive attention, while equally consequential questions surrounding accessibility, environmental sustainability and institutional accountability remain comparatively underdeveloped. As generative AI becomes embedded in teaching, learning and research, the gap between controlling technological misuse and governing its broader consequences is becoming increasingly significant.

A new international study, "Artificial Intelligence in Higher Education—Ethical Policies Mapping," published in the journal Education Sciences, investigates this imbalance. Authored by Kalin Penev, Vladislav Slavov and Radoslav Deliyski, researchers affiliated with Southampton Solent University in the United Kingdom and the Technical University of Sofia in Bulgaria, the study examines AI governance policies across 55 universities worldwide, complemented by a broader analysis of 87 institutional and related policy documents.

The findings reveal a clear hierarchy of institutional priorities. While 97.70% of analyzed documents include provisions addressing honesty and 96.55% address privacy and data protection, just 9.20% fully incorporate environmental sustainability. Accessibility, equitable participation, cybersecurity and responsible technology procurement also receive less consistent attention.

Academic Integrity Dominates as Universities Build Their AI Rulebooks

The expansion of generative AI has created a fundamental governance challenge for higher education. Technologies capable of generating written assignments, explaining complex concepts, producing research summaries and supporting personalized learning can enhance educational processes while complicating established expectations surrounding authorship, originality and independent intellectual work.

To understand institutional responses, the researchers examined publicly accessible university policies collected between February 2025 and May 2026. The institutional sample included 17 universities in Europe, 13 in Asia, nine in North America, six in Africa, five in South America and five in Oceania. A separate analysis of 87 documents incorporated institutional guidance alongside governmental, regional, professional and other relevant sources.

Across this broader document collection, ethical governance overwhelmingly emphasizes academic responsibility. Honesty appears as a fully included principle in 97.70% of documents, followed by privacy and data protection at 96.55%, academic integrity at 95.40%, accountability and ownership at 93.10%, and transparency at 91.95%. These figures describe the analyzed documents rather than the global university population.

The pattern reflects an institutional effort to distinguish legitimate technological assistance from academic misconduct. University policies commonly require students to remain responsible for submitted work, acknowledge AI-generated contributions and comply with assessment-specific instructions. The study highlights Imperial College London's treatment of unauthorized AI-generated assessment submissions as potential misconduct and Stellenbosch University's disclosure requirements for postgraduate research.

However, the researchers identify considerable variation in how policies translate broad ethical expectations into operational rules. Guidance frequently depends on individual courses, instructors and assessment requirements, creating situations where acceptable AI use may differ across academic settings. The central governance difficulty involves maintaining educational integrity without treating every form of technological assistance as inherently inappropriate.

The Missing Ethics: Accessibility, Sustainability and Institutional Risk

The study exposes substantially weaker attention to issues affecting the wider educational environment. Although university policies frequently establish expectations concerning honesty and responsibility, their treatment of equitable access and institutional technology management remains inconsistent.

Accessibility appears as a fully included principle in 63.22% of the 87 documents, while equity and non-discrimination reach 67.82%. Collaboration receives full inclusion in 45.98%. These findings indicate that concerns surrounding inclusive participation and coordinated governance have not achieved the same prominence as academic integrity and transparency.

Environmental sustainability presents the most pronounced gap. Only 9.20% of documents fully include sustainability, while 65.52% do not include it. The researchers argue that this omission deserves attention because AI systems depend on computational infrastructure associated with energy consumption, water requirements and other resource demands.

The contrast is particularly revealing because professional engineering and computing ethics frameworks frequently recognize environmental responsibility. University AI guidance, however, remains more narrowly concentrated on immediate educational conduct. The authors consequently advocate integrating sustainability into institutional policies rather than treating it as a separate technological concern.

Operational risks reveal another uneven area. Guidance warning users about cybersecurity threats, including deepfakes and phishing, is fully included in 57.47% of documents. Recommendations to consult institutional IT departments before procuring AI tools appear in 55.17%, indicating comparatively limited formal attention to technology acquisition and institutional oversight.

Privacy safeguards are more consistently developed. The study finds that 83.91% of documents fully include guidance against entering sensitive or confidential information into public AI systems, while 81.61% include data-minimization recommendations. These provisions demonstrate recognition that AI governance extends beyond plagiarism, although broader cybersecurity and procurement considerations remain less comprehensively addressed.

The findings suggest that universities have established stronger rules governing individual academic behavior than frameworks addressing the institutional responsibilities accompanying AI deployment. Comprehensive governance requires attention to technological access, system security, resource consumption and the conditions under which universities adopt AI services.

Universities Accept AI as a Learning Tool, but Struggle to Define Its Boundaries

Despite widespread concern over misuse, the research does not portray higher education as uniformly resistant to artificial intelligence. Instead, it identifies growing acceptance of AI applications that supplement learning without replacing students' intellectual engagement.

Among the 87 documents, 83.91% fully recognize supplementary learning as a legitimate AI application. Routine preparatory work appears in 73.56%, ideation and creativity in 65.52%, and personalized feedback in 64.37%. Such applications can help students explore unfamiliar concepts, generate preliminary ideas, receive explanations and prepare for more demanding academic activities.

The distinction between assistance and substitution runs throughout the analyzed guidance. AI-generated explanations, feedback and preparatory material may support educational objectives, while submitting machine-generated work without authorization or acknowledgment can undermine academic integrity. The acceptability of particular applications therefore depends on their purpose, disclosure requirements and relationship to intended learning outcomes.

This approach places considerable responsibility on students and educators to evaluate AI-generated material critically. The researchers emphasize that machine-produced responses can appear coherent without necessarily being accurate or demonstrating human-like understanding. Verification, independent judgment and the preservation of essential intellectual capabilities remain central to responsible educational use.

Nevertheless, institutional recognition of emerging applications remains uneven. Only 40.23% of documents fully include AI-supported active learning, while language practice receives full inclusion in 10.34%. Virtual and augmented reality applications appear in just 9.20%, suggesting that policy development remains concentrated on familiar generative tools rather than the wider technological possibilities associated with AI.

The study also identifies a conceptual weakness underlying university governance. Terms including intelligence, knowledge, reasoning and learning are not always precisely distinguished when describing AI systems. The authors argue that clearer terminology could reduce misunderstandings about machine capabilities and establish more reliable expectations regarding authorship, responsibility and educational decision-making.

From Compliance to Responsible Innovation: The Next Test for Higher Education

By examining ethical principles, practical guidance and recognized AI applications across different geographical and linguistic contexts, the study identifies the distance between formal commitments to responsible technology and the broader requirements of educational governance.

For university administrators, the findings support a more integrated policy approach. Academic integrity provisions remain essential, but institutions also need clearer procedures covering accessibility, technology procurement, cybersecurity, environmental responsibility and regular policy revision. The research places particular emphasis on aligning institutional governance with established regulatory frameworks and professional standards.

The authors discuss the European Union's General Data Protection Regulation and AI Act as relevant regulatory reference points. They also identify international AI management and data-quality standards as potential resources for developing clearer institutional responsibilities and more consistent governance procedures.

Education policymakers and development institutions can draw lessons from the uneven treatment of inclusion. Where students have differing access to technological resources, rules governing AI-assisted learning need to consider whether participation is genuinely equitable. The study does not measure these disparities directly, but its findings establish accessibility as a comparatively neglected dimension of formal policy development.

Environmental considerations introduce an additional strategic responsibility. The researchers recommend incorporating resource efficiency and sustainability awareness into university AI policies, linking responsible technological behavior with the professional values students may carry into future careers. Such an approach would extend AI ethics beyond immediate classroom conduct.

Important limitations nevertheless qualify the findings. The universities were selected through purposive rather than random sampling, preventing statistically representative regional comparisons. The analysis also excludes non-public institutional guidance, while multilingual interpretation and researcher judgment may affect the classification of policy provisions.

Most importantly, the study examines written governance documents rather than actual institutional practices. A policy's inclusion of transparency, accessibility or sustainability does not establish effective implementation, compliance or improved educational outcomes. The researchers therefore recommend longitudinal investigations, more balanced institutional samples, stronger multilingual validation and empirical assessments involving students, educators and administrators.

Future research must also examine how institutions translate ethical principles into enforceable procedures while adapting to emerging AI capabilities. Questions surrounding assessment design, responsible procurement, sustainability and access will become increasingly relevant as educational applications extend beyond conventional text-generation tools.