Artificial Intelligence in Higher Education—Ethical Policies Mapping

Abstract

Artificial Intelligence is transforming numerous sectors, and education is no exception. The adoption of generative AI tools in higher education presents significant opportunities for enhancing teaching, learning, assessment, accessibility, and administrative processes. At the same time, their widespread use raises critical ethical concerns related to academic integrity, bias, transparency, data protection, and the possible erosion of critical thinking and creativity. In response, higher education institutions worldwide are beginning to formulate ethical guidelines to govern the use of AI by students, educators, and administrators. However, institutional responses remain diverse across different contexts. This study examines how academic institutions across different countries conceptualise ethical AI use, define risks, and operationalise principles in practical policies through mapping 55 higher education institutions, complemented by an extended analysis of 87 documents. Across the six analysed thematic categories, transparency and disclosure, academic integrity, data protection and privacy, accountability, and risk mitigation are more consistently addressed than accessibility and inclusion. The analysis identifies practical guidance and recognised legitimate AI uses. The discussion considers the interpretation of AI and recommends aligning its use with ethical standards. The article concludes with a summary of the results and directions for future research.

1. Introduction

In a digital environment shaped by billions of connected devices, human–computer interaction has evolved from early forms of computer-mediated communication (Card et al., 1983; Hill et al., 2015), through neural network algorithms (Krizhevsky et al., 2012) and machine learning (LeCun et al., 2015), to contemporary large language models and artificial intelligence tools (Luo et al., 2025). The number and variety of AI tools are rapidly increasing, supported by growing computing resources, databases, and cloud infrastructure. These tools process text, images, media, code, and other forms of digital content. Examples include ChatGPT (2025), GLM-5.3-Flash (2026), BgGPT (2025), Llama (2025), IBM (2025), DeepSeek (2025), Qwen (2025), Doubao (2025), Kimi (2025), and Claude Opus 4.1 (2025). Their use is expanding across many sectors, and education is among the areas most significantly affected. At the same time, the adoption of generative AI tools in higher education presents significant challenges while also offering opportunities to enhance teaching, learning, assessment, admissions, enrolment, financial aid, course planning and degree progression, guidance, multilingual Q&A, and accommodation and accessibility support (Michel-Villarreal et al., 2023). According to the statistics presented in Figure 1 (Statista, 2023), ChatGPT, as an AI-powered chatbot, appears to have been most widely used by companies in the technology and education industries in 2023. An evaluation of the impact of ChatGPT on education (Rejeb et al., 2024) suggests that ChatGPT can offer educational advantages for both students and educators. It can contribute to more inclusive and personalised educational experiences, but institutions need to develop guidelines and policies for AI use in education. Among the growing number of AI concepts and solutions, multilingual large language models are particularly relevant. These models are trained on text from many languages, enabling them to process, understand, and generate content in different languages. Examples of these models include GPT-4o (OpenAI, 2024), Gemini (2025), Llama 3 (2025), Falcon LLM (2025), and Mistral AI (2025).

Figure 1. Number of companies using ChatGPT in their business in 2023, by industry (Statista, 2023).

Recent research has shown that higher education institutions are increasingly developing policies and guidelines to regulate, support, and pedagogically integrate generative AI, although institutional responses remain uneven across contexts and regions (Chan, 2023; Jin et al., 2025; McDonald et al., 2025; Moorhouse et al., 2023; Spivakovsky et al., 2023; Ullah et al., 2024). AI systems based on LLMs in native languages can drive AI adoption in various institutions, particularly in education.

Although AI tools can support personalised learning, facilitate accessibility, and streamline administrative workflows, their widespread use raises critical ethical concerns, including risks related to academic integrity, bias, transparency, data protection, and the erosion of essential human capabilities such as critical thinking and creativity (Jafari & Keykha, 2024; James Madison University Libraries, 2025). AI ethics can be understood as a set of values, principles, and techniques that employ widely accepted standards such as the IEEE Code of Ethics (Institute of Electrical and Electronics Engineers, 2020), the ACM Code of Ethics and Professional Conduct (Association for Computing Machinery, 2018), and the BCS Code of Conduct (BCS, 2022) to guide moral conduct in the development and use of AI systems. Developing an appropriate understanding of AI ethics and safety will be essential for the responsible use of AI across different areas. A dedicated study (Leslie, 2019) analyses ethical concerns and provides ethical guidance for the responsible design and implementation of AI systems in the public sector.

Government Digital Service (2019) notes that AI ethics has emerged from the need to address individual and societal harms that may arise unintentionally from AI systems. The guidance identifies three main sources of involuntary harm: misuse, questionable design, and unintended negative consequences. These include the use of systems beyond their intended purpose, insufficient consideration of algorithmic bias and safety risks, and failure to anticipate adverse effects on individuals and communities. These considerations are also relevant to the development of policies, strategies, and practices for AI applications in education. Multilingual capabilities can broaden the utility of AI models; however, performance may vary across languages and linguistic contexts, creating a risk of divergent interpretation.

The field of AI ethics should aim to prevent or mitigate potential negative impacts by providing project teams with guidance on the values, principles, and techniques needed to produce ethical, fair, and safe AI applications. In response to these developments, higher education institutions worldwide are beginning to formulate ethical guidelines to govern the use of AI by students, educators, and administrators. However, institutional responses remain diverse, often fragmented, and unevenly formalised across different contexts. While some universities have adopted comprehensive policy frameworks emphasising transparency, accountability, and responsible innovation, others are still grappling with the challenges posed by generative AI in education (New South Wales Government, 2025a; Russell Group, 2023; Universidad de Chile, 2025; University of Geneva, 2025). The aim of this study is to identify, critically evaluate, and document formal approaches to the ethical governance of AI in higher education through an international document analysis, with the literature review serving as contextual background.

Mapping frequently observed governance approaches involves identifying policy provisions that align with ethical principles. The focus is on formal guidance concerning AI use, its anticipated benefits and challenges, and associated ethical considerations. The analysis does not assess implementation effectiveness or institutional performance.

1.1. Research Questions

Based on a critical evaluation of the existing literature and the aims of the present study, the following research questions (RQs) were derived:

• RQ1. Which ethical governance principles are most consistently formalised in higher education AI policies and guidelines?
• RQ2. How are practical recommendations for responsible and ethical AI use operationalised in institutional and regulatory documents?
• RQ3. Which uses of AI are most frequently recognised as legitimate in higher education?
• RQ4. Which areas of AI governance remain less consistently addressed across the analysed documents?
• RQ5. How can taxonomic and ontological distinctions in the conceptualisation of artificial intelligence inform the development of clearer and safer university AI policies?

2. Methodology and Research Design

Previous studies have analysed institutional GenAI policies and guidelines across selected universities and regions, demonstrating the value of cross-document comparison for identifying common governance patterns, gaps, and institutional priorities (Jin et al., 2025; McDonald et al., 2025; Ullah et al., 2024). The first four research questions are addressed through the empirical analysis reported in Section 3, whereas the fifth is addressed through a conceptual synthesis of definitions and terminology derived from the specialised literature, regulatory standards, professional codes, and institutional sources, as discussed in Section 5. These two analytical strands are complementary: the empirical analysis identifies recurring governance patterns and gaps, while the conceptual synthesis examines how definitions and terminology may influence the interpretation and development of AI policies in higher education.

This study employs a qualitative comparative document analysis to examine institutional governance policies regulating the use of artificial intelligence in higher education following a structured multi-step workflow:

(1)

Potentially relevant documents were identified through an iterative and purposive search process. Documents were identified through major web search engines, direct searches of university and institutional websites, university policy repositories, and governmental and regional education policy sources. Additional potentially relevant policies and guidelines were identified through professional discussions, academic meetings, and exchanges with colleagues; these sources were subsequently included only when an official, publicly accessible document could be located and verified.

(2)

For the institution-level corpus used in Tables 3 and 4, documents were screened according to predefined inclusion criteria: recognised higher education institution, official policy or guideline status, public accessibility, relevance to AI use in higher education, and availability within the defined temporal scope.

(3)

Documents that did not meet these institution-level criteria, including informal commentaries, media publications, draft materials, and documents without explicit relevance to AI governance in education, were excluded from the institution-level corpus. A broader set of publicly accessible sources relevant to AI governance and ethical AI use in higher education was retained for the extended document-level analysis of core ethical principles, general guidelines for ethical AI use, and examples of legitimate AI use presented in Tables 5–7.

(4)

The selected documents were coded according to six analytical categories: transparency and disclosure, academic integrity, data protection and privacy, accountability, accessibility and inclusion, and risk mitigation.

(5)

The coding was cross-checked and verified to ensure consistency of interpretation across documents.

(6)

Two related coding schemes were used for different analytical purposes. The institution-level mapping in Tables 3 and 4 uses a binary threshold indicator: 1 indicates that a strong recommendation or mandatory provision is present, whereas 0 indicates that this threshold is not met. Because the purpose of this mapping is to provide a concise institution-level overview of whether each category reaches the strong or mandatory threshold, partial and absent cases are grouped together at this stage. For the institution-level mapping in Tables 3 and 4, the institution was the unit of analysis. Where more than one eligible policy or guidance document was available for the same institution, the documents were considered together in order to determine the institution-level classification for each analytical category. The extended document-level analysis in Tables 5–7 uses the three-level classification “included”, “partially included”, and “not included” to preserve greater analytical detail.

(7)

The quantitative mapping was interpreted through comparative synthesis to identify common ethical principles, practical guidelines, legitimate uses of AI, and remaining gaps in institutional AI governance.

The research was conducted in 2025–2026 and focuses on identifying, comparing, and interpreting publicly accessible university policies addressing the ethical, pedagogical, and regulatory use of AI in academic contexts. The methodological approach structures the analysis around three complementary dimensions: literature review, regulatory contextualisation, and comparative analysis of institutional governance approaches to artificial intelligence in higher education (McDonald et al., 2025). The study also draws on established approaches to the development of evaluation criteria in educational research (Jiang, 2016; Williams, 1983), adapting this logic to the analysis of AI policy documents. This allows the study to address questions that have not been fully explored in earlier research, particularly the international analysis of institutional AI governance frameworks (Bittle & El-Gayar, 2025; Ullah et al., 2024).

The analysis applies a qualitative comparative approach to a diverse set of publicly available documents addressing the ethical and regulatory dimensions of AI in academia. Six analytical criteria are used: transparency and disclosure, academic integrity, data protection and privacy, accountability, accessibility and inclusion, and risk mitigation. Building on the idea that both divergence and convergence can support adaptive analytical frameworks (Penev, 2022), the study intentionally includes institutions, regions, and languages from diverse geographical contexts. This diversity supports the identification of recurring ethical principles, regulatory approaches, and governance patterns across different institutional contexts. The broad methodological steps and inclusion criteria were established at the outset to support transparency and analytical consistency throughout the mapping process.

2.1. Corpus Construction and Selection Criteria

The study intentionally spans diverse institutions, geographical contexts, and languages to capture variation in ethical principles, practical guidelines, and models of institutional governance. The corpus was constructed through an iterative purposive search rather than through a predefined systematic database search. The search used Google, Bing, and Google Scholar, together with direct searches of university websites, institutional policy repositories, and relevant governmental and regional sources.

The discovery process was also informed by professional exchanges and academic meetings, which occasionally identified additional institutional or regulatory documents for subsequent verification. The search was progressively expanded across institutions, countries, and languages during the data-collection period from February 2025 to May 2026. The search concluded at the end of May 2026.

No formal saturation criterion was applied because the AI policy landscape was continuously evolving, with new documents being published and existing documents being revised, replaced, or withdrawn. Search terms were adapted iteratively and included combinations of terms such as “artificial intelligence”, “generative AI”, “AI policy”, “AI guidelines”, “ethical AI”, “university”, and “higher education”, together with institution- or country-specific terms where appropriate. No language restriction was applied; non-English documents were eligible for inclusion and were processed using the multilingual procedure described in Section 2.2. The selection criteria for the institution-level corpus are summarised in Table 1.

Table 1. Selection Criteria for the Institution-Level Corpus.

Criterion	Description
Institutional status	Recognised higher education institution
Policy availability	Official AI governance policy or guideline
Accessibility	Publicly available on institutional website
Formal status	Official institutional document
Temporal scope	Publicly available during February 2025–May 2026

Documents not meeting the institution-level criteria were excluded from Tables 3 and 4. A separate 87-document corpus was used for Tables 5–7, covering publicly accessible institutional, governmental, regional, sectoral, professional, and other relevant sources on AI governance in higher education.

A minimum of five institutions from each geographical area was used as a sampling criterion to support geographical diversity. The selected sample comprises 55 universities: 17 from Europe, 9 from North America, 5 from South America, 13 from Asia, 5 from Oceania, and 6 from Africa. The sampling strategy reflects a pragmatic analytical research design that prioritises transparency, accessibility, and institutional document analysis.

2.2. Analytical Procedure

The analysis followed a structured qualitative coding process.

First, an initial analytical framework, broadly corresponding to the six thematic categories used in the final analysis, was defined before the full coding process.

Second, the collected publicly accessible documents were reviewed and compared against this initial framework. The categories were subsequently refined through iterative comparison across institutional documents to improve their clarity, scope, and consistency of application.

Third, the refined thematic categories were applied through cross-document synthesis, enabling the identification of shared regulatory approaches, governance priorities, and structural similarities across institutional frameworks.

The final thematic structure presented in the Results section therefore reflects both the initial analytical framework and its subsequent refinement through cross-document comparison.

Three researchers participated in the document collection, coding, and verification process. The corpus was not coded independently in full by each researcher; instead, coding decisions and interpretations were cross-checked within the research team, and cases requiring clarification were discussed until agreement was reached. No formal intercoder reliability coefficient was calculated.

No separate coder-training or pilot-coding exercise was conducted. All three researchers had prior experience and subject-matter expertise relevant to the analysis. Coding was guided by the shared analytical framework, and consistency was supported through continuous communication, cross-checking, and discussion of cases requiring clarification. For documents not available in English, machine-assisted translation was used where necessary, with the original-language text consulted when clarification of terminology or meaning was required. Particular attention was given to terms whose meaning could vary across languages, especially where such differences could affect coding decisions.

Table 2 presents the thematic categories and typical policy elements used in the comparative analysis of selected policies for AI use in higher education.

Table 2. Main Analytical Thematic Categories.

Analytical Category/Criterion	Typical Policy Elements
Transparency and disclosure	Declaration of AI use in assignments and research
Academic integrity	Rules regarding plagiarism, misuse of AI tools
Data protection and privacy	Compliance with data protection frameworks
Accountability	Responsibility of students, faculty, and institutions
Accessibility and inclusion	Support for equitable access to AI tools
Risk mitigation	Measures addressing bias, misuse, and ethical risks

Figure 2 illustrates the analytical process applied to the institution-level corpus, beginning with the identification of potentially relevant documents, followed by screening according to the inclusion criteria described above. The documents are then coded according to the six analytical categories, followed by cross-checking and verification. The process concludes with quantitative aggregation and interpretive synthesis.

Figure 2. Analytical process.

The institution-level results are presented in Table 3 and Table 4. Transparency and disclosure are represented by C1, academic integrity by C2, data protection and privacy by C3, accountability by C4, accessibility and inclusion by C5, and risk mitigation by C6. For these tables, a binary threshold indicator is used: 1 indicates that a strong recommendation or mandatory provision is present, whereas 0 indicates that this threshold is not met. For more than one document from a single university, a logical disjunction applies. A value of 0 should therefore not be interpreted as complete absence, since it may also include partial or non-mandatory treatment.

Table 3. Mapping of Thematic Categories for Europe (N = 17 institutions).

University	Country	Region	C1	C2	C3	C4	C5	C6
Universidad de València	ES	Europe	1	1	1	1	0	1
Universidad de Navarra	ES	Europe	1	1	1	1	0	1
Wrocław University of Science and Technology	PL	Europe	1	1	1	1	0	1
Poznan University of Technology	PL	Europe	1	1	1	1	0	1
Vitebsk State University	BY	Europe	1	1	1	1	0	1
Politehnica University of Bucharest	RO	Europe	1	1	1	1	1	1
Philipps-Universität Marburg	DE	Europe	1	1	1	1	0	1
Bauhaus University Weimar	DE	Europe	1	1	1	1	0	1
University of Lübeck	DE	Europe	1	1	1	1	0	1
Università degli Studi di Bergamo	IT	Europe	1	1	1	1	0	1
University of Toulouse	FR	Europe	1	1	1	1	0	1
University of Lucerne	CH	Europe	1	1	1	1	0	1
Imperial College London	GB	Europe	1	1	1	1	0	1
Newcastle University	GB	Europe	1	1	0	1	0	1
University of Portsmouth	GB	Europe	1	1	0	1	0	1
Southampton Solent University	GB	Europe	1	1	1	1	1	1
Swansea University	GB	Europe	1	1	1	1	1	1

Table 4. Mapping of Thematic Categories for Asia, North and South America, Oceania, and Africa (N = 38 institutions).

University	Country	Region	C1	C2	C3	C4	C5	C6
Hefei University of Technology	CN	AS	1	1	1	1	0	1
Nankai University	CN	AS	1	1	1	1	0	1
Shanghai Jiao Tong University	CN	AS	1	1	1	1	0	1
ShanghaiTech University	CN	AS	1	1	1	1	0	1
Hong Kong Metropolitan University	CN	AS	1	1	1	1	0	1
Atilim University	TR	AS	1	1	1	1	0	1
Hacettepe University	TR	AS	1	1	1	1	1	1
Al-Farabi Kazakh National University	KZ	AS	1	1	1	1	0	1
Tomsk State University	RU	AS	1	1	1	1	0	1
Ryukoku University	JP	AS	1	1	1	1	0	1
Sophia University	JP	AS	1	1	1	1	0	1
Tokyo City University	JP	AS	1	1	1	1	0	1
Korea University	KR	AS	1	1	1	1	0	1
Universidad de Guanajuato	MX	NA	1	1	1	1	0	1
Universidad Abierta y a Distancia de México	MX	NA	1	1	1	1	0	1
Western University	CA	NA	1	1	1	1	1	0
Cornell University	US	NA	1	1	1	0	1	1
Elon University	US	NA	1	1	1	0	0	1
North Carolina State University	US	NA	1	1	1	1	0	1
Harvard University	US	NA	1	1	1	1	1	1
James Madison University	US	NA	0	0	1	0	1	1
St. John’s University	US	NA	1	1	1	1	1	1
Universidad Nacional Mayor de San Marcos	PE	SA	1	1	1	1	0	1
Universidad de Chile	CL	SA	1	1	1	1	0	1
Universidade de São Paulo	BR	SA	1	1	1	1	0	0
Universidade Estadual de Campinas	BR	SA	1	1	1	1	0	1
Universidad de los Andes	CO	SA	1	1	1	1	0	1
The University of Melbourne	AU	OC	1	1	1	1	1	1
Western Sydney University	AU	OC	1	1	1	1	1	1
Southern Cross University	AU	OC	1	1	1	1	0	1
Australian National University	AU	OC	1	1	1	1	0	1
University of Auckland	NZ	OC	1	1	1	1	0	0
North-West University	ZA	AF	1	1	1	1	0	1
University of Cape Town	ZA	AF	1	1	0	1	0	1
University of Pretoria	ZA	AF	1	1	1	1	0	1
University of Johannesburg	ZA	AF	1	1	0	1	1	1
Stellenbosch University	ZA	AF	1	1	1	1	0	1
Mohammed VI Polytechnic University	MA	AF	1	1	1	1	1	1

Note: AS = Asia; NA = North America; SA = South America; OC = Oceania; AF = Africa.

2.3. Regulatory Contextualisation

The institutional analysis was interpreted within the broader regulatory environment governing artificial intelligence and digital technologies in education. Particular attention was given to frameworks such as the General Data Protection Regulation (GDPR) and the European Union Artificial Intelligence Act (European Parliament & Council of the European Union, 2016, 2024). Regulatory analysis was not treated as a separate methodological strand but rather as a contextual layer supporting the interpretation of institutional governance approaches. Overarching regulatory frameworks were used for contextual interpretation, whereas selected governmental, regional, sectoral, professional and other relevant guidance documents were included in the extended document-level corpus (N = 87) when they were relevant to AI governance or ethical AI use in higher education and were coded as analytical units. This contextualisation supported the interpretation of institutional provisions alongside broader regulatory principles such as data protection, transparency, accountability, and risk management.

2.4. Evaluative Criteria for Identifying Promising Policy Provisions

The study applied explicit evaluative criteria to compare policy provisions and institutionally documented governance approaches. Policy provisions were assessed according to five criteria: normative coherence and clarity; alignment with existing regulatory frameworks; enforceability and operational specificity; protection of academic integrity and student rights; and balance between innovation support and risk mitigation. These criteria guided the analytical comparison of institutional policies and enabled the identification of governance approaches characterised by greater conceptual clarity, regulatory alignment, and operational specificity. Importantly, the criteria were not used to rank institutions but to support the comparative interpretation of different governance approaches.

3. Results

This section summarises the results of the policy analysis. Table 3 and Table 4 present an institution-level mapping of 55 higher education institutions. Table 5, Table 6 and Table 7 present a separate document-level analysis based on a corpus of 87 publicly accessible sources relevant to AI governance and ethical AI use in higher education, including institutional policies and guidelines, governmental and regional guidance, sectoral and professional frameworks, and other relevant source documents. In these tables, the individual document is the unit of analysis, and all percentages are calculated using the 87 analysed documents as the denominator. Each document contributed one observation to the document-level analysis, and no additional weighting was applied by institution, document type, or geographical origin. Documents are classified as “included”, “partially included”, or “not included”. “Included” denotes mandatory or required provisions, whereas “partially included” denotes recommended or suggested but non-mandatory provisions. The presentation of the results is aligned with RQ1–RQ4: Section 3.2 and Section 3.6 address RQ1, Section 3.7 addresses RQ2, Section 3.8 addresses RQ3, and Section 3.3 highlights the less consistently addressed areas relevant to RQ4.

3.1. International and Multilingual Scope of the Institutional Policy Corpus

The institutional sample spans diverse geographical and linguistic contexts. This international and multilingual scope is intended to support the identification of recurring patterns in the formal treatment of ethical AI governance rather than to provide representative regional comparisons.

3.2. Distribution of the Six Ethical Governance Categories

The mapping indicates that strong or mandatory policy provisions are most consistently observed in transparency and disclosure, academic integrity, data protection and privacy, accountability, and risk mitigation. The category of accessibility and inclusion is less consistently represented at the same level, suggesting that this dimension remains less explicitly formalised in many of the institutional documents examined.

3.3. Areas of Stronger and Weaker Policy Development

The institution-level analysis indicates more consistent policy coverage of academic integrity, transparency, privacy, accountability, and risk control. Accessibility and inclusion are less consistently addressed. The extended document-level analysis in Table 5, Table 6 and Table 7 also identifies weaker coverage of sustainability, cybersecurity, and responsible procurement.

3.4. Ethical and Unethical Use of AI

Imperial College London (2025a) provides the following examples of ethical and unethical uses of AI, particularly generative AI: ethical use includes using AI tools to support one’s own learning during the course, whereas unethical use includes using AI tools to generate, correct, or otherwise improve all or part of an assessment submission, except where such use is explicitly permitted. Unauthorised or inappropriate use of generative AI may constitute a breach of academic integrity or academic misconduct, including plagiarism or related forms of misconduct, depending on the applicable institutional rules (Australian National University, n.d.; Cornell University, 2023, 2025; Elon University, 2024, 2025; Tokyo City University, 2024; Universidad Abierta y a Distancia de México, 2025; Universidad de Guanajuato, 2025; Universidad de los Andes, 2024; Universidad de Navarra, 2024; Universidad de València, 2024; Universidade Estadual de Campinas, 2025, 2026; University of Auckland, 2026; University of Cape Town, 2025; University of Pretoria, 2026; Wrocław University of Science and Technology, 2026).

Some institutions also emphasise transparency through explicit disclosure requirements. Stellenbosch University, for example, requires postgraduate students to declare the use of AI tools in research assignments, theses, and dissertations (Stellenbosch University, 2025).

3.4.1. Plagiarism

Imperial College London (2025a) treats the unauthorised or unacknowledged submission of AI-generated work as academic misconduct, which may fall under plagiarism, collusion, or contract cheating. Comparable restrictions on the unauthorised or undisclosed submission of AI-generated work are reflected in the guidance of other institutions (Hong Kong Metropolitan University, 2025; Korea University Library, 2025; Mohammed VI Polytechnic University, 2025; Newcastle University, 2025; Ryukoku University, 2025; ShanghaiTech University, 2025; The University of Melbourne, 2025a, 2025b; Tomsk State University, 2024; Universidad de Chile, 2025; Universidad de València, 2024; Università degli Studi di Bergamo, 2025; University of Geneva, 2025).

3.4.2. Honesty and Integrity

Academic integrity and assessment design remain among the most prominent concerns in the literature on generative AI in higher education, particularly in relation to authorship, transparency, assessment validity, and misconduct (Bittle & El-Gayar, 2025; Ilieva et al., 2025; Moorhouse et al., 2023). Students should demonstrate a commitment to independence, honesty, and transparency. They should be honest and truthful in their dealings with each other and with third parties. Students engaged in research activities should conduct their research in a way that supports public trust and confidence in their institution’s research methods and findings. They should demonstrate rigour, honesty and integrity, and abide by relevant ethical and legal standards. Students should not become complicit in any activities in which a student gains an unfair advantage through plagiarism, self-plagiarism, collusion, examination offences, dishonest practice, or other means.

Students should comply with the applicable institutional policies on academic integrity and academic misconduct (Atilim University, 2025; Bauhaus University Weimar, 2024; Imperial College London, 2025a; North Carolina State University, 2023; Politehnica University of Bucharest, 2025; Universidad Nacional Mayor de San Marcos, 2025; Universidade de São Paulo, n.d.; University of Geneva, 2025; University of Johannesburg, 2024; Western University, 2026).

3.5. Risks and Issues to Consider When Using Generative AI Tools

Generative AI can produce fluent and apparently coherent responses to complex prompts. However, such outputs are generated through statistical modelling of language and should not be interpreted as evidence of human-like understanding (University of Portsmouth, 2025).

As a result, many institutions identify significant risks associated with the use of AI in higher education (Beijing Academy of Educational Sciences et al., 2024; Korea University Library, 2025; Philipps-Universität Marburg, 2023; Shanghai Jiao Tong University, 2025; Sophia University, 2025; University of Geneva, 2025; University of Lübeck, 2025; University of Lucerne, 2025).

3.6. Principles for the Use of AI in Universities

Concerns about the use of AI in higher education have led some universities to define basic principles for the ethical use of AI in the educational process. The results presented in Table 5, expressed as percentages of the total number of analysed documents, indicate a strong convergence across the analysed documents around a core set of ethical principles for AI use in higher education.

The most consistently represented principles are honesty, privacy and data protection, academic integrity, accountability and ownership, transparency, clear guidance, and compliance with institutional rules. This suggests that the analysed documents primarily frame ethical AI use through the lenses of responsible conduct, academic trust, and regulatory compliance. The high level of inclusion of privacy and data protection, academic integrity, honesty, transparency, and accountability demonstrates that the analysed sources are strongly oriented toward preventing misuse and preserving the credibility of academic work. These findings are consistent with the broader purpose of the study, namely, to identify how higher education institutions operationalise ethical principles in formal AI policies and guidelines.

At the same time, Table 5 below also shows that some principles remain less consistently represented across the analysed documents. Accessibility, equity and non-discrimination, collaboration, and especially sustainability appear less frequently or are often only partially addressed. This indicates that many AI policy and guidance documents have moved quickly to address the immediate risks of generative AI, such as plagiarism, non-disclosure, and data protection, but have not yet fully integrated broader systemic concerns. These include equal access to AI tools, inclusive pedagogical design, interdisciplinary governance, and the environmental implications of AI use. The comparatively weak presence of sustainability is particularly important. Although AI systems depend on substantial computational resources, environmental responsibility is still rarely treated as a core principle in the analysed AI policy and guidance documents. This suggests a gap between general professional codes of ethics, which often include responsibility toward society and the environment, and the more immediate operational focus of university AI guidelines.

Overall, these findings show that the ethical governance of AI in higher education is most consistently represented in relation to integrity, privacy, transparency, and accountability. However, the findings also suggest the need for a second stage of policy development, in which institutions move beyond compliance-focused guidance and address broader educational, social, and environmental dimensions of responsible AI use. The growing use of AI in higher education has prompted the development of guiding principles for its ethical application in teaching, learning, research, and administration that reflect a commitment to responsible innovation, student empowerment, and alignment with institutional missions.

Table 5. Core Ethical Principles for the Use of AI in Higher Education Identified in the Analysed Documents (N = 87 documents).

Principle	Description and Interpretation	Not Included	Included	Partially Included
AI Literacy	Support staff and students to understand, evaluate, and use generative AI tools critically and ethically.	3.45%	82.76%	13.79%
Clear Guidance	Communicate expectations about AI use in different academic settings.	4.60%	88.51%	6.90%
Compliance with Rules	Know and follow institutional AI policies, which may change over time.	5.75%	88.51%	5.75%
Lifelong Learning	Adopt a growth mindset for continuously evolving AI tools and practices.	2.30%	86.21%	11.49%
Human Abilities	Preserve empathy, creativity, and uniquely human skills alongside AI use.	0.00%	74.71%	25.29%
Privacy and Data Protection	Protect sensitive data, follow GDPR, and use secure platforms.	0.00%	96.55%	3.45%
Academic Integrity	Use AI transparently and honestly to uphold fairness and responsibility.	0.00%	95.40%	4.60%
Honesty	Be honest about the use of AI in all submitted work.	0.00%	97.70%	2.30%
Transparency	Declare when and how AI tools were used in coursework or research.	4.60%	91.95%	3.45%
Accountability and Ownership	Take responsibility for submitted work, including AI-generated content.	1.15%	93.10%	5.75%
Ethical Teaching and Assessment Design	Adapt learning outcomes and assessments to consider responsible AI use.	4.60%	78.16%	17.24%
Collaboration	Encourage interdisciplinary and cross-sector collaboration on AI in education.	20.69%	45.98%	33.33%
Dynamic Policy Review	Regularly evaluate AI guidelines and adapt them to new technologies and ethical issues.	4.60%	83.91%	11.49%
Equity and Non-discrimination	Ensure AI tools promote inclusion and do not disadvantage any groups.	4.60%	67.82%	27.59%
Accessibility	Guarantee equitable access to AI tools and resources.	12.64%	63.22%	24.14%
Policing and Penalties	Apply clear rules and penalties for inappropriate or undeclared AI use.	2.30%	83.91%	13.79%
Sustainability	Use AI responsibly in alignment with institutional sustainability goals.	65.52%	9.20%	25.29%

Table 5. Core Ethical Principles for the Use of AI in Higher Education Identified in the Analysed Documents (N = 87 documents).

Principle	Description and Interpretation	Not Included	Included	Partially Included
AI Literacy	Support staff and students to understand, evaluate, and use generative AI tools critically and ethically.	3.45%	82.76%	13.79%
Clear Guidance	Communicate expectations about AI use in different academic settings.	4.60%	88.51%	6.90%
Compliance with Rules	Know and follow institutional AI policies, which may change over time.	5.75%	88.51%	5.75%
Lifelong Learning	Adopt a growth mindset for continuously evolving AI tools and practices.	2.30%	86.21%	11.49%
Human Abilities	Preserve empathy, creativity, and uniquely human skills alongside AI use.	0.00%	74.71%	25.29%
Privacy and Data Protection	Protect sensitive data, follow GDPR, and use secure platforms.	0.00%	96.55%	3.45%
Academic Integrity	Use AI transparently and honestly to uphold fairness and responsibility.	0.00%	95.40%	4.60%
Honesty	Be honest about the use of AI in all submitted work.	0.00%	97.70%	2.30%
Transparency	Declare when and how AI tools were used in coursework or research.	4.60%	91.95%	3.45%
Accountability and Ownership	Take responsibility for submitted work, including AI-generated content.	1.15%	93.10%	5.75%
Ethical Teaching and Assessment Design	Adapt learning outcomes and assessments to consider responsible AI use.	4.60%	78.16%	17.24%
Collaboration	Encourage interdisciplinary and cross-sector collaboration on AI in education.	20.69%	45.98%	33.33%
Dynamic Policy Review	Regularly evaluate AI guidelines and adapt them to new technologies and ethical issues.	4.60%	83.91%	11.49%
Equity and Non-discrimination	Ensure AI tools promote inclusion and do not disadvantage any groups.	4.60%	67.82%	27.59%
Accessibility	Guarantee equitable access to AI tools and resources.	12.64%	63.22%	24.14%
Policing and Penalties	Apply clear rules and penalties for inappropriate or undeclared AI use.	2.30%	83.91%	13.79%
Sustainability	Use AI responsibly in alignment with institutional sustainability goals.	65.52%	9.20%	25.29%

The 87 documents analysed in Table 5, Table 6 and Table 7 are listed separately under “References to analysed documents” in Supplementary Materials.

Table 5 highlights recurring principles such as responsible AI-supported learning and future-oriented skills (Académie de Paris, 2024; Al-Farabi Kazakh National University, 2024; Atilim University, 2025; Beijing Academy of Educational Sciences et al., 2024; Gomel State University, 2025; Harvard University Information Technology, 2025; Hisar Okulları, 2025; Imperial College London, 2025b; Korea University Library, 2025; Russell Group, 2023), critical understanding of generative AI (University of Geneva, 2025; University of Portsmouth, 2025), and respect for human rights, privacy, data protection, and academic values (Crue Universidades Españolas, 2024; Deutscher Bundestag, 2025; Hacettepe University, 2025; Hefei University of Technology, 2025; James Madison University, 2025; Ministerio de Justicia Argentina, 2023; Poznan University of Technology, 2024; Southern Cross University, 2026; Swansea University, 2025a; Universidad de Chile, 2025; Vitebsk State University, 2025). Despite differences in wording, the analysed sources converge on themes of awareness, academic integrity, transparency, data protection, adaptability, and equity, as synthesised in Table 5.

3.7. Guidelines for the Use of AI in Universities

With the increasing presence of generative AI in academic settings, higher education institutions have issued practical guidance to support responsible use by students and staff. These recommendations address issues such as authorship, data security, fairness, academic honesty, and compliance with course-specific policies.

Table 6. General Guidelines for Ethical AI Use in Higher Education Identified in the Analysed Documents (N = 87 documents).

Thematic Area	Condensed Guideline	Not Included	Included	Partially Included
General Use	Do not use AI if you would not delegate the task to another person.	66.67%	14.94%	18.39%
General Use	Avoid uncritical use—AI output may lack depth and originality.	4.60%	37.93%	57.47%
General Use	If use is not explicitly allowed in assessments, it is likely not acceptable.	11.49%	51.72%	36.78%
Syllabus & Course Policy	AI rules vary by course and instructor—read the syllabus and ask if unclear.	5.75%	79.31%	14.94%
AI for Learning Only	Use AI to support—not replace—your skill development.	4.60%	91.95%	3.45%
Privacy & Security	Never input personal, confidential, or sensitive data into public AI tools.	1.15%	83.91%	14.94%
Awareness	Review faculty/course guidance. Ask instructors for clarification.	8.05%	82.76%	9.20%
Accuracy & Fact-Checking	Always verify the accuracy of AI-generated content. Do not rely solely on it.	1.15%	73.56%	25.29%
Authorship & Responsibility	The work you submit must be your own. Disclose AI use when required.	1.15%	85.06%	13.79%
Citations & Acknowledgement	AI-generated content must be clearly cited if quoted.	0.00%	89.66%	10.34%
Misconduct	Unacknowledged use of AI may lead to disciplinary action under academic misconduct policies.	2.30%	80.46%	17.24%
Data Minimisation	Apply privacy-by-design; include only minimal personal info and anonymise where possible.	0.00%	81.61%	18.39%
IT & Institutional Tools	Contact IT before procuring AI tools.	18.39%	55.17%	26.44%
Cybersecurity	AI risk warnings—be cautious of deepfakes and phishing.	1.15%	57.47%	41.38%

Table 6. General Guidelines for Ethical AI Use in Higher Education Identified in the Analysed Documents (N = 87 documents).

Thematic Area	Condensed Guideline	Not Included	Included	Partially Included
General Use	Do not use AI if you would not delegate the task to another person.	66.67%	14.94%	18.39%
General Use	Avoid uncritical use—AI output may lack depth and originality.	4.60%	37.93%	57.47%
General Use	If use is not explicitly allowed in assessments, it is likely not acceptable.	11.49%	51.72%	36.78%
Syllabus & Course Policy	AI rules vary by course and instructor—read the syllabus and ask if unclear.	5.75%	79.31%	14.94%
AI for Learning Only	Use AI to support—not replace—your skill development.	4.60%	91.95%	3.45%
Privacy & Security	Never input personal, confidential, or sensitive data into public AI tools.	1.15%	83.91%	14.94%
Awareness	Review faculty/course guidance. Ask instructors for clarification.	8.05%	82.76%	9.20%
Accuracy & Fact-Checking	Always verify the accuracy of AI-generated content. Do not rely solely on it.	1.15%	73.56%	25.29%
Authorship & Responsibility	The work you submit must be your own. Disclose AI use when required.	1.15%	85.06%	13.79%
Citations & Acknowledgement	AI-generated content must be clearly cited if quoted.	0.00%	89.66%	10.34%
Misconduct	Unacknowledged use of AI may lead to disciplinary action under academic misconduct policies.	2.30%	80.46%	17.24%
Data Minimisation	Apply privacy-by-design; include only minimal personal info and anonymise where possible.	0.00%	81.61%	18.39%
IT & Institutional Tools	Contact IT before procuring AI tools.	18.39%	55.17%	26.44%
Cybersecurity	AI risk warnings—be cautious of deepfakes and phishing.	1.15%	57.47%	41.38%

An analysis of the published guidance and related source documents in the extended corpus identifies common recommendations that are summarised in Table 6 above. The list of analysed documents is provided separately after the main reference list.

Table 6 presents a thematic grouping of the analysed guidance and related source documents for ethical AI use in higher education, with emphasis on responsible practices, data protection, academic integrity, and compliance with course-specific policies. It translates the broader ethical principles identified in Table 5 into practical guidance for students, teachers, and institutions.

The most frequently included guidelines relate to using AI as support for learning rather than as a substitute for student effort, acknowledging AI-generated content, protecting personal and sensitive data, following course-specific rules, and verifying the accuracy of AI outputs. This indicates that the analysed documents not only formulate broader ethical principles but also provide operational guidance for everyday academic practice.

An important finding is the strong emphasis on the student’s responsibility for submitted work. Guidelines on authorship, disclosure, citation, and academic misconduct show that many of the analysed documents treat AI use as part of the broader academic integrity framework. In this sense, AI is not necessarily prohibited, but its use must be transparent, justified, and aligned with the learning objectives of the course. The approach allows universities to support innovation while maintaining fairness in assessment.

The results also show that privacy and data minimisation have become central elements of ethical AI use. The high level of inclusion of recommendations not to enter personal, confidential, or sensitive data into public AI tools reflects growing awareness that AI ethics is not limited to plagiarism and assessment. It also includes data governance, security, and institutional responsibility for protecting students, staff, and research participants.

However, several areas remain less consistently developed. Guidance related to procurement of AI tools, institutional IT approval, cybersecurity risks, deepfakes, phishing, and broader digital security is less frequently included or is often only partially addressed. This suggests that many policies are still primarily educational and assessment-oriented, while the infrastructural and cybersecurity dimensions of AI governance remain underdeveloped.

Another finding is that some general recommendations, such as “do not use AI if you would not delegate the task to another person,” appear in relatively few documents. This may indicate that institutions prefer more concrete and context-specific rules instead of broad ethical analogies. It also reflects the complexity of AI use in education, where acceptable use depends heavily on the course, assessment type, learning outcomes, and instructor expectations.

Across the analysed documents, ethical AI guidance is practical but uneven. The strongest areas are academic integrity, disclosure, privacy, and responsible learning support. The less consistently addressed areas include cybersecurity, institutional procurement, and consistent interpretation across courses and departments. This confirms the need for clearer, more harmonised, and regularly updated institutional guidance.

However, such guidelines are not always standardised across institutions, courses, or departments. Students and educators are often expected to interpret them in relation to local contexts, disciplinary norms, and specific assessment instructions.

This makes transparency, critical awareness, and proactive communication particularly important.

3.8. Examples of Ethically Appropriate Uses of AI Technologies

Table 7 presents a quantitative assessment of examples of AI use identified as legitimate in the analysed documents, expressed as percentages of the total number of analysed documents. The table identifies the types of AI use that the analysed documents most commonly present as legitimate in higher education. The strongest pattern is the acceptance of AI as a supplementary learning tool. Most documents support the use of AI for additional explanations, examples, preparatory work, feedback, ideation, and support for understanding course material. This pattern indicates growing acceptance of AI as part of the learning environment, provided that its use remains transparent, bounded, and aligned with educational objectives.

The high level of inclusion of supplemental learning, routine preparatory work, personalised feedback, and ideation suggests that AI is generally presented as acceptable when it supports rather than replaces human learning. In this sense, AI is treated as a tool for scaffolding, reflection, and enhancement, not as an autonomous producer of academic work. This distinction is important for interpreting legitimate AI use: legitimacy depends not only on the tool itself, but also on the purpose, level of disclosure, and relationship to the intended learning outcomes.

The data also show that more advanced or discipline-specific applications are less consistently represented. Active learning, language practice, and especially virtual and augmented reality appear less frequently or are only partially included. This may be explained by the fact that many higher education AI policy and guidance documents are still focused on generative text-based tools, especially chatbots and large language models, while multimodal, immersive, and adaptive AI applications are not yet fully reflected in institutional guidance.

Table 7. Examples of Legitimate AI Use in Higher Education Identified in the Analysed Documents (N = 87 documents).

Thematic Area	Examples of Legitimate Use of AI	Not Included	Included	Partially Included
Supplemental Learning	AI can be a supplementary tool to provide additional explanations, examples, and resources to enhance students’ understanding of course materials.	6.90%	83.91%	9.20%
Discussion Facilitation	AI can stimulate discussions by generating prompts, questions, or hypothetical scenarios that encourage critical thinking and class participation.	24.14%	52.87%	22.99%
Ideation and Creativity	Students can use AI to generate or improve creative content in tasks such as idea generation, outlines, prototyping, and creation of alternative renditions and formats for concepts or visual artefacts.	13.79%	65.52%	20.69%
Active Learning	Instructors can use AI to create engaging assignments in which students work in groups to conduct a comparative analysis of AI-generated content related to coursework and curricular content.	34.48%	40.23%	25.29%
Personalised Feedback	AI can provide personalised feedback on assignments, projects, or assessments, addressing individual students’ questions and concerns.	14.94%	64.37%	20.69%
Language Practice and Learning	AI can allow students to practise language skills, simulate real-world conversations, or engage in various learning exercises. When learning foreign languages, students can receive immediate feedback on their pronunciation, grammar, and vocabulary.	39.08%	10.34%	50.57%
Virtual and Augmented Reality	AI can enhance virtual reality (VR) and augmented reality (AR) experiences in education by creating immersive simulations and scenarios, allowing students to explore complex concepts in a more engaging manner.	77.01%	9.20%	13.79%
Routine Preparatory Work	AI can support routine preparatory work, allowing students to focus on higher-order thinking and analysis.	2.30%	73.56%	24.14%

Table 7. Examples of Legitimate AI Use in Higher Education Identified in the Analysed Documents (N = 87 documents).

Thematic Area	Examples of Legitimate Use of AI	Not Included	Included	Partially Included
Supplemental Learning	AI can be a supplementary tool to provide additional explanations, examples, and resources to enhance students’ understanding of course materials.	6.90%	83.91%	9.20%
Discussion Facilitation	AI can stimulate discussions by generating prompts, questions, or hypothetical scenarios that encourage critical thinking and class participation.	24.14%	52.87%	22.99%
Ideation and Creativity	Students can use AI to generate or improve creative content in tasks such as idea generation, outlines, prototyping, and creation of alternative renditions and formats for concepts or visual artefacts.	13.79%	65.52%	20.69%
Active Learning	Instructors can use AI to create engaging assignments in which students work in groups to conduct a comparative analysis of AI-generated content related to coursework and curricular content.	34.48%	40.23%	25.29%
Personalised Feedback	AI can provide personalised feedback on assignments, projects, or assessments, addressing individual students’ questions and concerns.	14.94%	64.37%	20.69%
Language Practice and Learning	AI can allow students to practise language skills, simulate real-world conversations, or engage in various learning exercises. When learning foreign languages, students can receive immediate feedback on their pronunciation, grammar, and vocabulary.	39.08%	10.34%	50.57%
Virtual and Augmented Reality	AI can enhance virtual reality (VR) and augmented reality (AR) experiences in education by creating immersive simulations and scenarios, allowing students to explore complex concepts in a more engaging manner.	77.01%	9.20%	13.79%
Routine Preparatory Work	AI can support routine preparatory work, allowing students to focus on higher-order thinking and analysis.	2.30%	73.56%	24.14%

The relatively low inclusion of AI-enhanced VR and AR should not necessarily be interpreted as rejection. Rather, it suggests that institutional AI policies are still at an early stage and tend to address the most visible and widely used tools first. As AI becomes embedded in simulation environments, adaptive learning platforms, language learning systems, and assessment technologies, future policies will need to expand their definitions of legitimate use beyond text generation.

Table 7 also highlights the importance of pedagogical framing. The same AI activity may be ethical or unethical depending on how it is used. For example, using AI to generate ideas, compare alternative explanations, or receive formative feedback can support learning. By contrast, submitting AI-generated output as one’s own work, using AI to bypass the development of required competences, or relying on AI without verification would conflict with the principles of academic integrity and accountability identified in Table 5.

Table 7 indicates that the educational potential of AI is recognised particularly in supportive and formative roles. The next challenge for higher education policy is to define more clearly how legitimate AI use can be embedded in curriculum design, assessment, feedback, accessibility, and lifelong learning without weakening student agency, authorship, critical thinking, and responsibility.

4. Limitations

This study has several limitations that should be considered when interpreting the findings. First, the corpus was constructed using a pragmatic and non-random sampling strategy. The number of institutions differs across geographical areas, and the study therefore does not claim statistical representativeness or population-level regional comparison. Future research could extend the corpus and apply a more balanced sampling design to support more systematic geographical comparison.

Second, the analysis is based only on publicly accessible source documents relevant to AI governance and ethical AI use in higher education. These include institutional policies and guidelines, governmental and regional guidance, sectoral and professional frameworks, and other relevant sources included in the extended corpus. Internal or non-public guidance could not be included. AI-related policies are also developing rapidly, and some documents may have been revised, replaced, or removed after the data-collection period. Longitudinal studies could help examine how institutional AI governance changes over time.

Third, the multilingual character of the corpus creates an additional interpretive limitation. Machine-assisted translation was used where necessary, and semantic differences across languages may affect the interpretation of terms such as “intelligence”, “learning”, “knowledge”, and related concepts, as well as binary coding in Table 3 and Table 4 where a value of 1 represents a mandatory provision or strong recommendation and a value of 0 combines cases in which the strong, mandatory threshold is not reached, potentially including both partial treatment and absence. Coder judgement therefore remains a possible source of subjectivity, even where coding was cross-checked. Future studies could strengthen this aspect through more systematic multilingual validation and formal intercoder reliability procedures.

Finally, the analysed documents are heterogeneous in type and scope, and written policy should not be treated as equivalent to actual institutional practice. The reported frequencies indicate how often particular themes are present in the analysed corpus, but they do not measure the quality, effectiveness, implementation, or compliance of the policies themselves. Further empirical work could examine how formal AI policies are interpreted and implemented by students, academic staff, and university administrators.

5. Discussion

The corpus-level findings generated by the document analysis are reported in Section 3. The present Discussion extends beyond these empirical results to consider broader conceptual and normative issues concerning how AI is defined and interpreted in higher education policy. Drawing on the literature, institutional definitions, regulatory standards, and professional codes, it discusses ontological and taxonomic aspects of artificial intelligence that may influence its ethical use and the development of university policies. These considerations are presented as contextual interpretations and recommendations rather than as findings generated by the coded corpus. Artificial intelligence encompasses a broad range of systems and applications capable of performing tasks commonly associated with human intelligence, including language processing, content generation, and decision support (Oxford English Dictionary, 2023). In higher education, AI may also support accessibility and personalised learning, including opportunities for remote learners and students with disabilities (James Madison University Libraries, 2025). This broad range of applications makes the way AI is defined and conceptualised in institutional policies particularly relevant to its ethical governance. Across different domains and publications, various definitions of artificial intelligence, generative artificial intelligence, and ethical AI use can be found. Although many institutions understand these concepts in similar ways, important differences remain in terminology, emphasis, and practical interpretation.

A common understanding of AI is that it is a technology that enables machines to imitate various complex human skills (Sheikh et al., 2023). However, potentially misleading interpretations can be found in some popular sources and university policies. UNESCO similarly describes AI systems as potentially capable of imitating or exceeding certain human cognitive capacities, including sensing, language interaction, reasoning, problem solving, and creativity (Sabzalieva & Valentini, 2023). Many of the policies examined use concepts such as “intelligence”, “learning”, “reasoning”, “knowledge”, and “content”, although the degree of conceptual precision and differentiation between these terms varies across documents.

The University of Lincoln draws on the Oxford English Dictionary definition of AI as the capacity of machines to exhibit or simulate intelligent behaviour and to perform tasks previously associated with human intelligence (Oxford English Dictionary, 2023; University of Lincoln, 2025). Western Sydney University, drawing on the New South Wales AI Assurance Framework, describes AI in terms of advanced computational methods that identify meaningful patterns in data and support decision-making (New South Wales Government, 2025b; Western Sydney University, 2025). St. John’s University similarly characterises AI as involving machines that perform tasks commonly associated with human intelligence through data-driven learning, pattern recognition, natural language processing, and interaction with the environment (St. John’s University, 2024). AI may also include tools for research summarisation, automated data analysis, and synthetic data generation (Bekker, 2024). As AI tools continue to develop rapidly, definitions of AI should not be treated as fixed, and emerging forms of AI should not be excluded (North-West University, 2024). Buckley and Zwarthoed (2024) conceptualise AI as the simulation of human intelligence processes by computer systems, including learning, reasoning, prediction, content generation, and self-correction. This emphasis on simulation helps distinguish artificial from natural intelligence.

Generative AI is a specific type of AI capable of generating new content such as text, images, music, and more. It integrates technologies such as natural language processing, machine learning, and deep learning and is trained on large volumes of data. For example, tools such as ChatGPT (2025), Gemini (2025), GLM-5.3-Flash (2026), BgGPT (2025) and similar tools can answer questions, provide explanations and summaries, draft documents, and simulate discussions, among other things. Other examples of AI technologies include voice dictation and speech recognition, computer vision, robotics, recommender systems, and knowledge graphs (St. John’s University, 2024).

Western Sydney University adopts the New South Wales Government’s broad definition of generative AI as artificial intelligence capable of producing new content, including text, images, video, audio, or code (New South Wales Government, 2025a; Western Sydney University, 2025).

Harvard University and North-West University similarly describe generative AI as systems that learn from and mimic large amounts of data to create different forms of content in response to inputs or prompts (Harvard University Information Technology, 2025; North-West University, 2024).

An examination of authoritative, normative, and conceptual definitions and interpretations reveals inconsistencies in how artificial intelligence is understood (American Psychological Association, 2018; Chinese Dictionary, n.d.; Copeland, 2026; Dudenredaktion, n.d.; Institute for Bulgarian Language, n.d.; McCarthy et al., 2006; Oxford English Dictionary, 2023; Priberam, n.d.; Real Academia Española, n.d.; Vassilev et al., 2024, 2025). Attempts to translate these concepts across languages may further increase the risk of misunderstanding and divergent expectations (DeepL Translator, 2026).

From a different perspective, the definition of AI proposed by the High-Level Expert Group on Artificial Intelligence (2019) raises questions about its suitability for policy development. This definition leaves room for further clarification regarding its scope and operational interpretation.

A clearer distinction between data, information, content, knowledge, and responsibility could help improve university AI policies. Many institutional documents use terms such as ‘learning’, ‘reasoning’, ‘knowledge’, and ‘intelligence’, but these terms are not always clearly defined. This may create unrealistic expectations about what AI systems can do and may blur the distinction between human cognitive performance and machine-based simulation. For this reason, future AI policies in higher education should use more precise terminology when describing the role of AI in learning, assessment, authorship, and decision-making.

The rapid growth in the number of artificial intelligence tools and applications raises questions about resource use, leading to global ethical issues and risks. University policies on the ethical use of artificial intelligence in education could benefit significantly from following widely accepted standards and codes of ethical conduct. An analysis of professional codes of ethics and conduct from scientific, engineering, and computing associations reveals a recurring emphasis on environmental responsibility. These codes commonly encourage professionals to support sustainable development, disclose risks that may threaten the public or the environment, and adopt environmentally responsible practices in the design and use of technology. Taken together, they frame the protection of people, society, and the natural environment as an ethical responsibility and encourage professionals to minimise the negative impacts associated with technological development and use (Association for Computing Machinery, 2018; BCS, 2022; Brazilian Computer Society, 2024; China Computer Federation, 2023; Código Ético y Deontológico de la Ingeniería Informática, 2026; Institute of Electrical and Electronics Engineers, 2020; Russian Union of Scientific and Engineering Associations, 2002).

Establishing environmental protection as a fundamental principle in university policies on the use of artificial intelligence could have two positive consequences:

• Reducing irresponsible AI use and the unnecessary consumption of resources by students and staff;
• Enhancing students’ awareness and understanding of the ethical use of artificial intelligence in their future professional careers.

In this sense, published studies and statistical data indicate that the energy consumption of artificial intelligence systems requires ethical consideration (Luccioni et al., 2024; Maslej et al., 2024). The water footprint of AI models also deserves attention (Li et al., 2025). These issues should be reflected in regulatory standards, institutional frameworks, and legislation.

ISO/IEC 42001:2023, an international standard that specifies requirements for an Artificial Intelligence Management System (AIMS), could support the development of ethical AI policies in education for entities providing or utilising AI-based products or services, thereby supporting the responsible development and use of AI systems (International Organization for Standardization, 2023). Certification requirements for AI providers could also inform university ethical policies. The design, implementation, and continuous improvement of AI policies would benefit from clear and consistent definitions and interpretations of AI. Several recurring conceptual issues can be identified across definitions published in the literature (Osborn, 2023; Oxford English Dictionary, 2023; Vassilev et al., 2024, 2025). A first factor concerns the understanding and interpretation of the term intelligence. In cross-language translation, ‘intelligence’ and ‘intellect’ may be conflated (DeepL Translator, 2026), which warrants particular attention in educational institutions responsible for developing knowledge and understanding.

A second factor concerns the gap between what AI is perceived to be capable of and what AI systems can actually do. Many definitions are based on the assumption that AI can perform tasks or exhibit behaviour similar to humans (Oxford English Dictionary, 2023; Vassilev et al., 2024, 2025). However, since AI systems process data through computational models, policies should clarify the distinction between simulation, modelling, and genuine human performance.

A third factor concerns the type of material on which artificial intelligence operates. AI systems process digitised text, large language model outputs, datasets stored in data centres, and other forms of digital content. University policies should distinguish such digital content from human knowledge and, where relevant, from wisdom. This distinction matters for reliability, trustworthiness, currency, accuracy, level of abstraction, and the time needed for correction and validation. The ISO/IEC 5259 series on data quality for analytics and machine learning may provide useful guidance (International Organization for Standardization, 2024).

A fourth factor concerns the way AI systems are improved. Many definitions refer to machine learning, which is implemented through different types of neural networks and related algorithms whose efficiency may vary (Chen, 2024).

The identification of ethical risks therefore requires a clearer distinction between human learning and the computational adjustment of neural network parameters through preprogrammed operations. Finally, according to Article 6 of Regulation (EU) 2024/1689 (European Parliament & Council of the European Union, 2024), an AI system may be considered high-risk when specific conditions are fulfilled. Applying this risk-based perspective to university policies could support safer and more secure use of AI in higher education.

With regard to RQ5, the conceptual arguments discussed suggest that clearer taxonomic and ontological distinctions between human cognitive concepts and machine-based operations can support more precise definitions of AI capabilities, responsibilities, and risk boundaries in university policies.

Examining the issues discussed above regarding definitions and interpretation could help in developing more precise policies for the use of AI in higher education.

6. Conclusions

In summary, this article presents findings from a structured international comparative document analysis of ethical dimensions of AI use in higher education.

It examined similarities and variations across six thematic categories through which the ethical use of artificial intelligence is conceptualised, risks are defined, and principles are translated into practical policies in different institutional contexts. The study also considered regulatory frameworks, such as the General Data Protection Regulation (GDPR), the EU AI Act, and codes of ethics and professional conduct, as contextual reference points for interpreting institutional AI governance. In comparison to earlier publications (Bittle & El-Gayar, 2025; Ullah et al., 2024), the study takes a slightly different focus and highlights the following contributions:

• It includes an international and multilingual corpus of university policies and related documents;
• It links university policies to the regulatory context and practical guidelines;
• It employs six analytical categories: transparency and disclosure, academic integrity, data protection and privacy, accountability, accessibility and inclusion, and risk mitigation;
• It extends the scope of the analysis by including not only ethical principles but also operational guidelines and examples of legitimate AI use;
• It highlights both more consistently addressed areas—such as privacy, transparency, and academic integrity—and less consistently addressed areas, such as accessibility, sustainability, cybersecurity, responsible procurement, and conceptual clarity regarding “exhibiting human-like behaviour” versus “machine-based simulation”.

Drawing on a structured comparative analysis of institutional and related policy and guidance documents, supported by the relevant research literature, the study (i) synthesised existing definitions and ethical frameworks; (ii) identified recurrent risks and approaches; and (iii) documented how the analysed sources describe appropriate and inappropriate uses of AI in academia.

Further research should examine how higher education institutions can move from general ethical principles toward more operational, enforceable, and pedagogically meaningful AI governance frameworks. Future studies could further investigate how universities define and distinguish concepts such as intelligence, learning, reasoning, content, knowledge, and responsibility in AI-related policies. Particular attention should be given to the distinction between human cognitive performance and machine-based simulation, as this distinction has important implications for academic integrity, authorship, assessment design, and student responsibility. Further research may also explore how institutional AI policies can better integrate emerging standards for AI management, data quality, cybersecurity, sustainability, accessibility, and responsible procurement of AI tools.