AI Governance in Higher Education: Balancing Innovation, Data Security and Regulatory Compliance
Abstract:
Artificial Intelligence (AI) is rapidly transforming higher education in various ways and presenting opportunities to enhance teaching, learning experiences, research productivity, and administrative efficiency. Despite the significant potential of AI in higher education, its adoption raises important concerns relating to data privacy, cybersecurity, algorithmic bias, transparency and regulatory compliance. AI governance is essential in higher education to ensure alignment with institutional values and regulatory standards. This paper examines the governance implications of AI adoption in higher education through a qualitative and exploratory approach based on a review of the literature and an analysis of relevant regulatory frameworks including the General Data Protection Regulation (GDPR) and European Union Artificial Intelligence Act (EU AI Act) 2024. This approach was used to identify recurring governance concerns and inform the development of a conceptual framework. The findings of this paper reveal seven interrelated governance challenges: data privacy and re-identification risks, dependence on third-party vendors, misinformation and data integrity issues, cybersecurity vulnerabilities, algorithmic bias, inadequate consent mechanisms, and governance, accountability and regulatory uncertainty. Based on the findings, the study proposes a three-pillar conceptual framework that brings together AI innovation, governance challenges, and governance and compliance mechanisms. Drawing on Responsible Innovation Theory, Information Privacy Theory, and Institutional Theory, the conceptual framework illustrates how higher education institutions can pursue the benefits of AI while addressing security, ethical, and regulatory concerns. The framework may also provide a useful reference for institutions operating in different regulatory contexts, particularly where AI governance continues to evolve. The framework is conceptual and has not yet been empirically validated.1. Introduction
Artificial intelligence (AI) has emerged as one of the most significant technological developments that is transforming all aspects of higher education. AI technologies such as personalised learning platforms, intelligent tutoring systems, predictive analytics and text generative AI tools are being integrated throughout the higher education system at an accelerated rate (Ocaña-Fernández et al., 2019; Walczak & Cellary, 2023). It is becoming inevitable that institutions will need to leverage this emerging technology in order to improve pedagogical practices and administrative efficiencies, as well as support research capabilities (Pratama et al., 2023). Hence, AI is becoming a key element in many modernization initiatives being adopted by higher education institutions.
While AI offers significant opportunities to transform many aspects of higher education, its adoption is also associated with numerous risks. Many of the newer generation of AI-based applications require substantial amounts of sensitive personal and institutional data to operate efficiently. In higher education, this creates substantial vulnerabilities with regard to the protection of student and institutional data as well as potential breaches of confidentiality (Huang, 2023). Furthermore, the use of AI-based systems increases the risk of cyber-attacks against institutional systems by adding another layer of vulnerability created through the use of third-party service providers, complex AI architectures, and evolving sophisticated cyber threats directed toward educational platforms (Alhitmi et al., 2024; Okulich-Kazarin et al., 2024).
Despite the growing body of literature on AI integration in higher education settings, there remains a critical gap in understanding how institutions can develop and implement effective governance models for AI-based solutions while ensuring strong data security protocols remain in place. Current research tends to focus either on the positive impacts of AI on pedagogy (Khare et al., 2018) or exclusively on the technical issues of data security without considering the larger governance structure in which those technical considerations exist (Afolabi, 2024). Recent frameworks demonstrate substantial progress in addressing the issue of developing effective AI governance models for the higher education sector (Chan, 2023; Shahid et al., 2026). At the international level, there are various frameworks such as United Nations Educational, Scientific and Cultural Organization (UNESCO)’s AI Ethics Recommendation and the Organisation for Economic Co-operation and Development (OECD) AI Principles that provide broader principles concerning responsible, human-centered, transparent, and accountable AI use (Organisation for Economic Co-operation and Development, 2023; United Nations Educational, Scientific and Cultural Organization, 2023). Although these frameworks set out important principles for responsible AI governance, higher education institutions still face challenges in converting them into practical governance arrangements that reflect their specific organizational needs and responsibilities.
Recent research indicates growing alignment between higher education governance practices and international AI principles. However, higher education institutions differ in how these principles are interpreted and implemented (Khoza et al., 2026). These variations in governance practices across institutions suggest that international alignment at the principal level does not necessarily translate into consistent implementation within higher education.
Despite the growing recognition of these governance requirements, existing research suggests that institutional governance has not always kept pace with the rapid development and adoption of AI technologies in higher education (Khoza et al., 2026). Current AI policies and guidance in education also tend to remain broad and implicit, reflecting the limited evidence on how AI governance principles are translated into institutional practice (Chan, 2023).
Therefore, more attention is needed to how these broad principles can be operationalized within the higher education context through an integrated approach that addresses innovation, data security, ethical accountability and regulatory considerations while taking into consideration the institutional structures, policies and decision-making processes (Uslu & Bozkurt, 2026).
The absence of such an integrated framework is particularly noticeable in Kuwait, where the rapid expansion of AI adoption in universities has outpaced the development of dedicated AI governance standards. In this paper, Kuwait is used as an illustrative example of a jurisdiction where AI adoption is expanding more rapidly than AI-specific regulatory development.
This research aims to address that gap by carrying out a narrative review of the relevant literature, conducting a regulatory evaluation of the European Union (EU)’s General Data Protection Regulation (GDPR) and the European Union Artificial Intelligence Act (EU AI Act) 2024 and developing a conceptual framework for the management of AI in higher education. The proposed conceptual framework positions legal compliance as a foundation rather than a ceiling for the governance of AI in higher education. Additionally, the conceptual framework illustrates how such internationally developed frameworks may be applicable to non-EU nations, specifically Kuwait, which has experienced increased adoption of AI-enabled technologies by higher education institutions in the absence of equivalent nationally-based regulatory frameworks.
This study is guided by three interrelated research questions addressing the opportunities, risks and governance implications of AI adoption in higher education.
• What are the principal AI technologies being integrated into higher education institutions, what benefits do they offer and what data security, privacy and ethical challenges do they create?
This question is concerned with examining the benefits and associated risks of AI adoption in higher education. This is essential for understanding the governance requirements needed to support responsible AI integration.
• What legal and regulatory frameworks govern the deployment of AI in educational settings?
This question focuses on examining regulatory instruments such as the GDPR and the EU AI Act 2024, which help to clarify the legal obligations applicable to educational institutions and provide useful reference points for AI governance beyond the EU.
• How can a conceptual framework be developed to support the responsible governance of AI in higher education institutions?
Building on the first and second research questions, this study seeks to propose a framework that integrates innovation objectives, governance challenges and regulatory requirements. The proposed framework is intended to provide an analytical tool for future research and practical guidance for institutional leaders and policymakers.
This research makes three main contributions to the literature. First, it provides a theoretically grounded analysis of AI governance in higher education by examining the relationship between AI innovation, data security, and ethical accountability. As a result, it responds to the gap between the rapid adoption of AI technologies and the development of institutional governance mechanisms that are designed to address the new risks and responsibilities associated with the use of AI within the higher education context.
Second, the study also provides a regulatory analysis of the GDPR and the EU AI Act 2024 in relation to the use of AI within higher education. This analysis helps clarify key regulatory requirements and demonstrates how these frameworks can be used as useful reference points and sources of guidance for institutions seeking to translate broad governance principles into more specific and practical institutional practices, regardless of whether they are based within or outside of the EU.
Third, the study proposes a three-pillar conceptual framework that integrates the concepts of AI innovation, governance challenges, and governance and compliance mechanisms. The proposed framework addresses the need for a more integrated institutional approach to governing AI in higher education. It also provides a structured basis for decision-makers, policymakers, and researchers to consider the interconnected technological, security, ethical, and regulatory dimensions of AI adoption within the higher education context.
The three-pillar framework differs from existing higher education AI governance models in three respects. First, Chan’s (2023) framework organises AI policy around pedagogical, governance and operational dimensions and Shahid et al. (2026) concentrate on generative AI and large language models. The present framework instead follows the sequence through which governance problems arise: it starts from the AI technologies an institution adopts; identifies the risks each adoption creates in a university setting. It links each risk to specific legal provisions and institutional safeguards. Second, it uses the GDPR and the EU AI Act as comparative benchmarks. Therefore, institutions in jurisdictions without AI-specific legislation, such as Kuwait, can measure their practices against detailed regulatory requirements rather than against general principles alone. Third, it treats governance as an iterative process in which problems found during monitoring feed back into decisions about how a system is designed and used. Another framework is needed because existing guidance tends to remain at the level of broad principles and institutions still lack a structure that connects particular AI applications to particular risks and responses. Section 8 compares the framework with existing approaches in more detail.
2. Theoretical Background
This research draws on three complementary theoretical perspectives to ground the conceptual framework and inform the analysis of AI governance in higher education.
Responsible Innovation Theory provides the primary normative foundation for this study. Developed by Stilgoe et al. (2013), the theory argues that technological innovation should not be evaluated solely on the basis of efficiency, productivity or economic value. Instead, innovation processes should be guided by larger societal considerations such as social accountability, ethical responsibility, and sustainability.
The theory developed by Stilgoe et al. (2013) outlines four connected dimensions of responsible innovation: anticipation, inclusion, reflexivity and responsiveness. The first dimension, anticipation, requires organizations to consider and think about the potential risks or consequences of an emerging technology prior to its implementation. For example, when evaluating the use of AI in higher education, these include assessing how these systems will affect students’ right to privacy, academic integrity, algorithmic bias toward certain groups and the level of student autonomy. The second dimension, inclusion, is concerned with the need for a wide range of stakeholders to participate in decisions made regarding AI systems so that different stakeholders, such as students, faculty members, administrators and technical staff, can have input in the governance process. The third dimension, reflexivity, emphasizes that institutions question whether their current assumptions are correct as well as ask what the ethical implications could be of using AI in making educational decisions (Aguilar et al., 2023). The fourth dimension, responsiveness, requires institutions to be adaptive and to continue to adjust the governance of AI systems with the development of new technologies and the emerging stakeholder concerns and regulatory expectations.
In the context of higher education, Responsible Innovation Theory emphasizes that AI governance needs to go further than just meeting technical requirements and managing risks. Universities must actively shape the development and deployment of AI technologies in ways that align with educational values associated with teaching and learning, academic freedom, fairness and social responsibility. From this perspective, responsible innovation provides the normative justification for the proposed governance framework, which seeks to embed ethical considerations into institutional AI decision-making processes. In addition, institutional governance, ethical oversight, and stakeholder engagement play an important role in determining how effectively AI adoption can support learning, innovation, and broader educational outcomes.
Information Privacy Theory offers a complementary perspective by focusing on protecting individuals’ personal data and individual autonomy. Westin’s (1967) foundational theory of information privacy conceptualises privacy as an individual’s right to manage or limit the amount of information about them that is collected, used, and distributed. This theoretical perspective has direct applicability to the AI environment in higher education. This is because there is substantial potential for violations of an individual’s privacy due to the extensive collection of student data (academic records, behavioral trends, biometric data, and educational engagement metrics).
Another useful taxonomy of privacy was developed by Clarke (1999), who distinguished four dimensions: bodily privacy, behavioural privacy, communication privacy, and informational privacy. Bodily privacy concerns the integrity of the person’s body, which becomes relevant where AI systems rely on biometric data such as facial images or fingerprints. Communication privacy is related to the flow of information between two parties, behavioural privacy is concerned with an individual’s actions monitored by another, and informational privacy is related to control over data about oneself and how that data is used. AI can potentially interfere with all aspects of privacy through automated monitoring, predictive analytics and algorithmic profiling. For instance, learning analytics platforms can create detailed behavioral profiles of students. Another example is AI-driven assessment systems that can collect vast amounts of personal and academic data without providing clear explanations of how decisions are made.
From this perspective, effective AI governance cannot be limited solely to compliance with data protection legislation. Effective AI governance will require institutions to implement mechanisms that ensure transparency, accountability, informed consent, data minimization and best practice in securing personal/academic data. Therefore, information privacy provides an important basis for understanding how higher education institutions should govern the collection, use, access, and protection of personal data and institutional data when adopting AI technologies.
Institutional Theory provides an organizational perspective to better understand how organizations adopt and implement AI governance practices in higher education. According to DiMaggio & Powell (1983), organisations often become increasingly similar over time as a result of institutional pressures that shape their behaviour. The three primary forms of institutional pressure that influence organizational behavior include coercive, normative and mimetic isomorphism. Coercive isomorphism is related to formal legal and regulatory obligations. In the context of AI governance, instruments such as the GDPR and the EU AI Act create external pressures that encourage institutions to adopt specific compliance measures. Normative isomorphism is related to professional standards, ethical guidelines and expectations within academic and professional communities (Abane & Cobbina, 2024). For example, universities may adopt AI governance practices not only because they are legally required to do so, but also because such practices are perceived as professionally appropriate and ethically necessary. Hence, there is both a coercive force to comply with these formal regulations, and a normative force to align with what is considered “best practice” based upon professional association guidelines. Mimetic isomorphism exists when institutions adopt models developed by peer organizations. This occurs most frequently in situations characterized by high levels of uncertainty and technological complexity (Abane & Cobbina, 2024).
Institutional Theory is considered relevant for understanding AI governance and can offer insights into how AI governance develops at institutions in environments without significant AI-specific regulatory oversight. In such environments, given that comprehensive AI regulation remains underdeveloped, coercive regulatory pressures are relatively limited. Thus, mimetic and normative isomorphic forces can drive much of the development of governance.
Higher education institutions may choose to establish AI governance policies based on international best practices, ethical considerations and/or professional standards to enhance institutional legitimacy, strengthen stakeholder trust and demonstrate responsible stewardship of new technologies. This can also help institutions to build public confidence in their ability to govern new technologies effectively (Uslu & Bozkurt, 2026).
Therefore, Institutional Theory explains how ethical AI governance practices can become embedded within organisational structures and decision-making processes, even in the absence of extensive legal mandates. It also shows that institutional norms, professional expectations, and stakeholder perceptions play a critical role in governing the use of AI technologies in higher education.
3. Methodology
This study employs an exploratory and qualitative research approach based on a narrative review of academic literature and a critical analysis of relevant legal and regulatory frameworks governing AI and data protection. The paper seeks to synthesise existing knowledge, identify major governance challenges associated with AI adoption in higher education and develop a conceptual framework to support institutional decision-making. A qualitative approach is well-suited for this study because it examines normative, ethical and governance issues that require contextual interpretation rather than quantitative measurement. The methodology is structured around three complementary stages designed to ensure analytical rigour, transparency and conceptual coherence.
The first stage involved a narrative literature review of peer-reviewed journal articles, books, institutional reports, policy documents, and international guidance relating to AI governance, data protection, ethical AI, and higher education. A narrative review provides a suitable approach for examining multidisciplinary topics by bringing together knowledge developed across different theoretical perspectives and fields of research (Snyder, 2019).
Relevant sources were identified through searches of Scopus, Web of Science, Google Scholar, and publisher databases. The search covered material published between 2018, the year in which the GDPR became applicable, and 2026. Foundational theoretical and methodological works published before this period (for example, Westin, 1967; DiMaggio & Powell, 1983; Stilgoe et al., 2013) were also included because they underpin the analytical approach of the study. Search strings combined three groups of terms with Boolean operators: (1) technology terms (“artificial intelligence” OR “generative AI” OR “machine learning” OR “learning analytics” OR “predictive analytics”); (2) context terms (“higher education” OR universit* OR “tertiary education”); and (3) governance terms (governance OR “data protection” OR privacy OR cybersecurity OR ethics OR regulation OR GDPR OR “AI Act”). Additional searches combined the context terms with “Kuwait” to locate national studies and official documents.
Sources were included if they were peer-reviewed articles, scholarly books or chapters, or official policy and regulatory documents published in English, and if they addressed AI governance, data protection, security or ethics either in higher education or in a setting whose findings could reasonably be applied to higher education. Studies concerned only with the technical design of AI systems, without a governance, privacy or ethical dimension, were excluded, as were duplicate records. Titles and abstracts were screened first, the full texts of potentially relevant sources were then read, and the reference lists of central sources were checked to identify further material. Because this is a narrative rather than a systematic review, the aim was to capture the main debates in the field rather than to retrieve every relevant publication.
This narrative review developed through an iterative process, and the number of records retrieved from each database at each screening stage was not recorded. This is a limitation of the study. The search is described using the Scale for the Quality Assessment of Narrative Review Articles scale for assessing the quality of narrative reviews, which asks authors to explain their search terms, inclusion criteria and the types of literature included (Baethge et al., 2019). The review drew on a range of sources, including peer-reviewed journal articles, academic books and book chapters, conference papers, legal and policy documents, institutional reports and news articles.
In particular, more attention was given to literature addressing responsible innovation, information privacy, institutional governance, algorithmic accountability, and emerging AI regulatory frameworks. The literature was critically examined and synthesized in order to identify recurring issues, governance challenges, and relevant practices associated with the adoption of AI technologies within higher education institutions. In addition, this stage also provided the theoretical and conceptual foundation for the study and informed the subsequent regulatory analysis.
The second stage consisted of a critical and comparative analysis of the principal legal instruments currently shaping AI governance and data protection. More specifically, the GDPR and the EU AI Act 2024 were analysed as they represent influential regulatory models for governing AI systems and the processing of personal data.
The analysis focused mainly on identifying provisions that are most relevant to the use of AI within higher education institutions, including requirements related to risk assessment, transparency, accountability, human oversight, data governance, and privacy protection. The authors focused on the classification of AI systems according to risk levels under the EU AI Act and the corresponding obligations imposed on institutions using these systems.
The provisions were selected by asking which rules would apply to three common university uses of AI: admissions and access decisions, student assessment (including online exam proctoring) and learning analytics used to identify students at risk. For each use, the analysis identified the GDPR articles on lawful processing, information to students, automated decision-making and impact assessment and the EU AI Act provisions on high-risk classification, deployer obligations and prohibited practices. The results are summarised in Section 5. The analysis also separated two kinds of rule. The first consists of rules that bind Kuwaiti universities directly, mainly Law No. 20 of 2014 on Electronic Transactions and for public institutions, the cybersecurity requirements of the National Cybersecurity Center (NCSC). The second consists of instruments that do not bind Kuwaiti universities but are used in this paper as governance references, the GDPR and the EU AI Act. Section 4.1 sets out this distinction.
The study also considers the broader implications of these regulatory frameworks for jurisdictions outside the EU. This is particularly relevant to Kuwait, where dedicated AI legislation remains limited but where universities increasingly engage with international technology providers operating under European regulatory standards. Consequently, the GDPR and AI Act provide useful reference points for examining emerging governance expectations within higher education institutions.
The third stage involved a critical synthesis of the literature review and regulatory analysis to inform the development of the proposed conceptual framework. In this stage, the authors combined the key governance concerns and recurring issues identified across the reviewed literature and regulatory materials.
The analysis focused on areas of convergence and connection across the reviewed sources. Following the logic of thematic synthesis (Thomas & Harden, 2008), recurring issues were first coded descriptively and then grouped into broader analytical themes, which produced the seven governance challenges discussed in Section 6. Recurring issues were considered in relation to broader areas including innovation, data security, privacy, accountability, transparency, and institutional governance. These areas were then critically examined to inform the development of the three-pillar conceptual framework.
The seven challenges were derived in four steps. First, passages in the retained sources that described a risk, harm or governance difficulty linked to AI use were extracted. Second, each passage was given a short descriptive code that stayed close to the wording of the source, such as “re-identification through linked datasets”, “fabricated references” or “students cannot opt out of required platforms”. Third, codes that described the same underlying governance problem were grouped, and each group was given an analytical label. Fourth, a group was treated as a separate challenge when it recurred across several independent sources and could be linked to at least one provision of the GDPR or the EU AI Act. Table 1 gives examples of the codes and the main sources behind each challenge.
Governance Challenge | Examples of Initial Codes | Main Sources |
Data privacy and re-identification risks (6.1) | Re-identification through linked datasets; limits of anonymisation; sensitive academic and behavioural data | Huang (2023); Majeed & Lee (2021); Pikhart & Al-Obaydi (2025) |
Third-party vendor dependence and data control (6.2) | Outsourced processing; little visibility into how vendors train models; vendor lock-in; cross-border transfers | Huang (2023); Jin et al. (2025) |
Misinformation, data integrity and AI-generated content (6.3) | Fabricated references; inaccurate outputs; unclear authorship and originality | Alhitmi et al. (2024); Gruenhagen et al. (2024); Walczak & Cellary (2023) |
Cybersecurity vulnerabilities and emerging attack vectors (6.4) | New attack surfaces; AI-assisted phishing; attacks on machine learning models | Binhammad et al. (2024); Nikkhah et al. (2024); Okulich-Kazarin et al. (2024) |
Algorithmic bias and automated decision-making (6.5) | Biased training data; opaque predictions; automated admission and progression decisions | Idowu et al. (2024) |
Inadequate consent mechanisms (6.6) | Long and complex privacy notices; compulsory platforms; secondary use of data | Cate & Mayer-Schönberger (2013); Susser (2019); Cohen & Slottje (2024); Nazaretsky et al. (2025) |
Governance, accountability and regulatory uncertainty (6.7) | Fragmented responsibility; policies lagging behind adoption; no national AI law | Alboloushi et al. (2026); Chan (2023); Khoza et al. (2026); Uslu & Bozkurt (2026) |
The resulting framework is intended to provide an analytical lens for understanding AI governance in higher education and a practical governance tool for policymakers, university leaders, and institutional decision-makers. By bringing together findings from the literature and contemporary regulatory developments, the proposed framework seeks to support the responsible adoption of AI technologies that ensures appropriate safeguards for data security, ethical accountability, and stakeholder trust.
As this paper is based on a narrative and critical review, the findings are dependent on the scope and quality of the literature and regulatory materials examined. Accordingly, the proposed framework should be regarded as a conceptual framework intended to guide future research and institutional practice rather than an empirically validated governance mechanism.
4. AI Integration in Higher Education: Benefits and Applications
AI is increasingly transforming higher education through its integration into teaching, learning, research and institutional administration. Universities are deploying a diverse range of AI technologies to improve educational outcomes, enhance operational and administrative efficiency and support decision-making. As AI capabilities continue to evolve, their adoption has expanded beyond isolated applications to become embedded across multiple dimensions of university operations.
Within the educational context, AI-powered personalised learning systems use machine learning algorithms to adapt instructional content, learning pathways and assessment activities to individual student needs (Pratama et al., 2023). Research suggests that such systems can improve student engagement and learning outcomes by accommodating differences in learning styles, pace and academic performance (Aldosari, 2020; Pratama et al., 2023). Similarly, intelligent tutoring systems provide real-time academic support and personalised feedback outside traditional classroom settings, extending access to educational assistance beyond conventional teaching hours.
Assessment practices have also been influenced by AI technologies. For instance, automated grading systems and AI-supported assessment tools can help in reducing the administrative burdens on academic staff while providing students with more timely feedback (Walczak & Cellary, 2023). In addition, predictive analytics systems are increasingly being used to detect students who may be considered at risk of academic failure or withdrawal (Khan et al., 2025). By analysing patterns within student data, these systems enable institutions to implement early intervention practices and strategies that support student retention and academic success.
Beyond teaching and learning, AI is increasingly shaping institutional administration (Agrawal et al., 2019). Universities employ AI-powered systems for admissions screening, student services, campus security, resource allocation and strategic planning. These applications enable institutions to automate routine processes, optimise decision-making and improve operational efficiency. Furthermore, AI-supported research tools have enhanced the ability of researchers to analyse large datasets, identify patterns and generate insights that would be difficult to obtain through traditional methods alone (Chauke et al., 2024).
Together, these developments demonstrate the significant benefits associated with AI adoption in higher education. These benefits include personalised learning at scale, improved administrative efficiency, enhanced research capabilities, greater access to academic support and improved student retention through data-driven interventions (Sodagari, 2025). As a result, AI is increasingly viewed as a strategic resource capable of supporting institutional innovation and competitiveness in an evolving educational landscape.
AI adoption within Kuwaiti higher education institutions has expanded significantly in recent years, which reflects the broader national digital transformation initiatives and regional ambitions to develop knowledge-based economies (Alboloushi et al., 2026). In Kuwait, AI-supported learning platforms, automated assessment systems and administrative technologies are increasingly being introduced across the education system, and national initiatives such as the Ministry of Education’s AI education project launched in cooperation with Google in 2026 show the policy priority given to AI in education (Kuwait News Agency, 2026).
Developments at Kuwait University, the country’s main public university, illustrate how institutions are responding. In November 2024, the University Council approved a Generative Artificial Intelligence (GenAI) Policy (Kuwait University, 2024). The policy allows faculty members to set the rules for GenAI use in their own courses, treats unauthorised use as a form of cheating or plagiarism, permits instructors to require students to disclose the tools and prompts they used and advises staff not to upload sensitive data to GenAI tools. It is a clear example of institution-level governance developed in the absence of a national AI law. Its focus, however, is on academic integrity and classroom use. It says little about how the university should assess, procure or monitor AI systems that process student data at scale, such as predictive analytics, automated assessment or admissions tools, which are the systems most likely to fall within the high-risk category of the EU AI Act.
Student uptake shows why this gap matters. In a survey of 179 education students in Kuwait, Almisad (2026) found that most respondents already held an account on at least one GenAI application, most commonly ChatGPT, and that students’ confidence in the privacy of these tools was associated with more favourable attitudes towards using them for learning. Adoption is therefore already widespread among students, while the privacy assurances on which they rely are shaped largely by external providers rather than by their institutions.
Despite this growth, Kuwait does not yet possess a comprehensive AI-specific regulatory framework (Alboloushi et al., 2026). As a result, institutions increasingly rely on internal governance measures and international best practices to manage emerging risks relating to privacy, security and accountability. As universities continue to adopt AI technologies, the absence of detailed domestic regulatory guidance places greater responsibility on institutions to develop mechanisms that ensure transparency, accountability, ethical oversight and data security.
Kuwait has no single, comprehensive data protection statute. The main general rules are in Law No. 20 of 2014 on Electronic Transactions, which applies to government authorities, public institutions and companies and therefore covers both public and private universities (State of Kuwait, 2014; Saleh et al., 2026). Under Articles 32 to 37 of that law, an entity that collects personal data must obtain consent and state the purpose of collection, keep the data accurate and protected, and not disclose it without consent. Data subjects may ask to see their data and to have it modified or deleted, and unlawful disclosure is a criminal offence. These rules apply to any AI system a university uses to process student or staff data, but they were not written with automated decision-making, profiling or algorithmic bias in mind. Public universities also fall under the NCSC, established by Decree No. 37 of 2022, which oversees cybersecurity in government entities (Saleh et al., 2026).
AI governance in Kuwaiti higher education should be considered within the national regulatory environment and the broader international governance landscape. At the national level, the Communication and Information Technology Regulatory Authority (CITRA) issued a Data Privacy Protection Regulation in 2021, which was replaced by Resolution No. 26 of 2024 (CITRA, 2021; CITRA, 2024). The 2024 regulation requires informed consent, clear privacy notices in Arabic and English, purpose limitation and notification of data breaches to CITRA within 24 hours. However, its obligations are limited to CITRA-licensed telecommunications and internet service providers (Saleh et al., 2026), so universities are affected only indirectly, for example when they rely on a licensed provider for cloud or connectivity services. In addition, the draft Kuwait National Artificial Intelligence Strategy 2025–2028 names education as a priority sector and proposes measures on AI governance, privacy and security, but it is a strategy document rather than binding AI legislation (State of Kuwait, 2025). At the international level, the OECD AI Principles and UNESCO’s Recommendation on the Ethics of Artificial Intelligence provide broader guidance on responsible AI, emphasizing principles such as transparency, accountability, fairness, human oversight, and the protection of individual rights (Organisation for Economic Co-operation and Development, 2023; United Nations Educational, Scientific and Cultural Organization, 2023).
Instrument | Scope | Status for Kuwaiti Universities |
Law No. 20 of 2014 on Electronic Transactions (Arts. 32–37) | Consent, purpose, accuracy, security, access, correction and deletion of personal data held by government bodies, public institutions and companies | Binding (applies directly) |
Decree No. 37 of 2022 establishing the NCSC | Cybersecurity oversight, particularly of government entities | Binding for public universities |
CITRA Data Privacy Protection Regulation (Resolution No. 26 of 2024) | Personal data processed by CITRA-licensed telecommunications and internet providers | Indirect: binds the university’s licensed providers, not the university |
Kuwait University GenAI Policy (2024) | Use of generative AI by Kuwait University students and staff | Binding within Kuwait University only |
Draft Kuwait National AI Strategy 2025–2028 | National priorities, including education and AI governance | Not binding (policy direction) |
GDPR (Regulation (EU) 2016/679) | Processing of personal data in the EU | Governance reference; applies directly only where a university offers services to, or monitors the behaviour of, individuals in the EU (Art. 3(2)) |
EU AI Act (Regulation (EU) 2024/1689) | Providers and deployers of AI systems placed on the EU market or whose output is used in the EU | Governance reference; applies to a non-EU deployer only where the system’s output is used in the EU (Art. 2(1)(c)) |
OECD AI Principles; UNESCO Recommendation on the Ethics of AI | International principles for trustworthy AI | Not binding (guidance) |
Table 2 summarises these instruments and shows which of them bind Kuwaiti universities directly and which are used in this paper only as governance references. Two points stand out. First, the binding rules that apply to universities are general data protection and cybersecurity rules, and none of them deals specifically with AI. Second, at the institutional level, Kuwait University’s GenAI Policy is used as the example in this paper because it is publicly available and was approved by the University Council. It governs how students and staff use generative AI tools, but it does not set out a procedure for approving or monitoring AI systems that the university itself deploys. The framework in Section 7 is designed to address that gap.
In addition, the GDPR and the EU AI Act provide influential regulatory benchmarks for data protection and risk-based AI governance. While acknowledging these broader national and international frameworks, this paper focuses primarily on the GDPR and the EU AI Act as comparative regulatory benchmarks because they provide detailed and established requirements for data protection, transparency, accountability, human oversight, and AI risk management. Their structured regulatory provisions provide a useful basis for critically examining the governance and data protection challenges associated with AI adoption in higher education within the Kuwaiti context.
5. Regulatory Frameworks Governing AI in Higher Education
The increasing adoption of AI in higher education has prompted the development of regulatory frameworks designed to address concerns relating to privacy, accountability, transparency and risk management. Within the EU, two instruments are particularly significant: the GDPR (European Parliament & Council of the European Union, 2016) and the EU AI Act 2024 (European Parliament & Council of the European Union, 2024). Together, these frameworks establish a comprehensive regulatory environment governing the collection and processing of personal data as well as the development and deployment of AI systems.
The GDPR provides the principal legal framework for the protection of personal data within the EU. Its provisions are particularly relevant to higher education institutions that increasingly rely on AI-driven systems to collect, analyse and process student and staff information. Its core principles, which include lawfulness, transparency, fairness, purpose limitation, data minimization, accuracy, storage limitation and accountability, establish important safeguards for educational data processing activities. The GDPR also imposes specific obligations relating to data subject rights, data security, breach notification and the use of automated decision-making systems, making it a foundational component of AI governance within educational environments.
Complementing the GDPR, the EU AI Act 2024 establishes the world’s first comprehensive regulatory framework specifically governing AI systems. The Act adopts a risk-based approach that classifies AI systems into four categories: unacceptable risk, high-risk, limited risk and minimal risk (Schuett, 2024). The level of regulatory scrutiny applied to an AI system depends upon its potential impact on individuals and society.
Many AI systems deployed within higher education, particularly those used for admissions, student assessment, behavioral monitoring and predictive analytics, may fall within the high-risk category. Such systems are subject to a range of obligations, including risk management requirements, data governance standards, technical documentation, transparency measures, human oversight mechanisms and requirements relating to accuracy, robustness and cybersecurity. The Act therefore extends beyond traditional data protection concerns and introduces a broader governance framework focused on the responsible development and deployment of AI technologies.
Annex III, point 3 of the EU AI Act lists four educational uses as high-risk: AI systems used to determine access or admission to educational institutions; to evaluate learning outcomes, including when those outcomes are used to steer a student’s learning process; to assess the level of education a person will receive or be able to access; and to monitor and detect prohibited behavior of students during tests. A system in these categories is not treated as high-risk under Article 6(3) only where it performs a narrow procedural or preparatory task and does not materially influence the outcome of a decision. A university that uses a high-risk system is a deployer, and Article 26 requires it to use the system in line with the provider’s instructions, assign human oversight to staff with the necessary competence, monitor the system’s operation, keep the logs it generates and inform the students concerned that the system is used to make or assist decisions about them. Where the university is a public body or a private body providing a public service, Article 27 also requires a fundamental rights impact assessment before deployment, and Article 86 gives affected students a right to an explanation of decisions based on the system’s output. Separately, Article 5(1)(f) prohibits AI systems that infer the emotions of people in educational institutions, except for medical or safety reasons, which rules out emotion-recognition features in proctoring and classroom-monitoring tools. The prohibitions have applied since 2 February 2025. Following the amendments made by Regulation (EU) 2026/1744, the obligations for high-risk systems listed in Annex III apply from 2 December 2027 (European Parliament & Council of the European Union, 2026). Table 3 relates these provisions, and the most relevant GDPR articles, to common university uses of AI.
The interaction between the GDPR and the EU AI Act creates a complementary regulatory structure. While the GDPR primarily governs the processing and protection of personal data, the EU AI Act focuses on the risks associated with AI systems themselves. Together, they provide a coherent framework that promotes transparency, accountability, human oversight and trust in AI-enabled decision-making.
Although neither instrument applies directly to many non-EU jurisdictions, including Kuwait, their influence extends beyond European borders. Universities increasingly procure AI services from international technology providers that operate within European regulatory environments and design their systems to comply with these standards. Consequently, the GDPR and the EU AI Act are likely to shape governance expectations and institutional practices even where equivalent domestic legislation does not yet exist. For higher education institutions in Kuwait, these frameworks provide valuable reference points for developing governance structures capable of supporting the responsible and secure adoption of AI technologies. In this paper, the two instruments are used as governance references rather than as applicable law with two narrow exceptions. The GDPR applies directly to Kuwaiti universities that offer educational services to, or monitors the behaviour of, individuals located in the EU (Article 3(2)) and the EU AI Act applies where the output of an AI system used by the university is used in the EU (Article 2(1)(c)). The rules that bind Kuwaiti universities directly are those marked as binding in Table 2.
University Use | Main GDPR Provisions | Main EU AI Act Provisions |
Admissions and access (e.g., screening or ranking applicants) | Art. 6 (lawful basis); Arts. 13–14 (information to data subjects, including applicants); Art. 22 (decisions based solely on automated processing; right to human intervention); Art. 35 (data protection impact assessment) | High-risk: Annex III, point 3(a); Art. 26 (deployer duties, including informing applicants, Art. 26(11)); Art. 27 (fundamental rights impact assessment); Art. 86 (right to explanation) |
Student assessment and online exam proctoring | Art. 5(1)(c) (data minimisation); Art. 9 (biometric data used to identify students); Art. 22; Art. 35 | High-risk: Annex III, points 3(b) and 3(d) Art. 5(1)(f) (ban on emotion recognition in education); Arts. 14 and 26 (human oversight) |
Learning analytics and early-warning systems | Art. 5(1)(b) (purpose limitation); Art. 6(1)(e) or (f) (public task or legitimate interests); Art. 21 (right to object); Art. 22 (where a flag triggers a decision without human review); Art. 35 (systematic monitoring) | High-risk where used to evaluate learning outcomes or steer learning: Annex III, point 3(b); Possibly outside the high-risk category under Art. 6(3) if only preparatory; Art. 4 (AI literacy of staff) |
Generative AI tools licensed for students and staff | Art. 28 (processor contracts); Arts. 44–49 (transfers outside the EU) | Art. 50 (transparency for chatbots and AI-generated content); Art. 4 (AI literacy) |
6. Data Security Challenges in AI-Driven Higher Education
While AI offers significant opportunities for innovation in higher education, its deployment also introduces a range of data security, privacy and governance challenges. The thematic synthesis conducted in this study (see Section 3.3) identified seven interrelated categories of risk that have emerged consistently across the literature. These challenges extend beyond technical security concerns and encompass broader questions relating to accountability, transparency, regulatory compliance and institutional trust. Understanding these risks is essential for developing governance frameworks capable of supporting the responsible use of AI in educational environments.
Although many of these risks also arise in sectors such as healthcare, finance and government, their manifestation in higher education is shaped by the distinctive relationship between universities, students and academic staff. Higher education institutions process extensive academic, behavioural and personal data; make decisions affecting admission, assessment, progression and student opportunities; and often require students to use institutionally selected digital platforms as part of their studies. These characteristics create particular concerns regarding student autonomy, academic integrity, institutional power asymmetries and the preservation of academic values. The discussion below therefore examines not only the general nature of these risks but also their specific implications within higher education.
Data privacy remains one of the most significant concerns associated with AI adoption in higher education (Pikhart & Al-Obaydi, 2025). AI systems frequently rely on large volumes of student and staff data, including academic records, behavioural information, learning analytics, biometric data and engagement metrics. Although institutions often employ anonymisation or pseudonymisation techniques, advances in machine learning have increased the possibility of re-identifying individuals through the combination of multiple datasets (Majeed & Lee, 2021).
This risk has particular significance in higher education because student information is often accumulated across multiple interconnected systems, including learning-management platforms, assessment tools, attendance systems, student-service databases and predictive analytics applications. When combined, these datasets may reveal detailed patterns concerning academic performance, behaviour and engagement that extend beyond ordinary administrative records. Re-identification may therefore affect not only informational privacy but also students’ academic reputation, educational opportunities and confidence in institutional digital systems.
The risk of re-identification raises important concerns regarding individual autonomy, informed consent and institutional accountability (Pikhart & Al-Obaydi, 2025). These concerns are reflected in Article 5(1)(b) and (c) of the GDPR, which establish the principles of purpose limitation and data minimisation. Under the EU AI Act, high-risk AI systems are subject to the requirements set out in Articles 8–15, including risk management, data governance, transparency, human oversight, accuracy, robustness and cybersecurity. Certain AI systems used in education fall within the high-risk categories identified under Article 6(2), read together with Annex III, point 3, subject to the conditions and exceptions set out in Article 6(3). These include systems used for admission or access to educational institutions and for evaluating learning outcomes. For higher education institutions, particularly in jurisdictions where AI-specific regulation remains limited, the possibility of unintended disclosure or re-identification highlights the need for robust governance mechanisms and privacy-by-design approaches.
The growing reliance on external AI providers presents another significant governance challenge. Universities increasingly procure AI-enabled platforms and services from third-party vendors, often transferring substantial volumes of institutional and personal data to external systems (Jin et al., 2025). While such arrangements provide access to advanced technologies and specialist expertise, they may simultaneously reduce institutional control over data processing activities (Huang, 2023).
The higher education context adds a further dimension to this dependency because students and faculty members may have little influence over the choice of institutional platforms. Students may, for example, be required to use a particular learning, assessment or administrative system in order to participate fully in their studies. This creates a power asymmetry in which institutional procurement decisions may expose students to external data-processing practices that they cannot realistically avoid.
Dependence on third-party providers creates risks relating to unauthorised data sharing, opaque processing practices, cross-border data transfers and long-term vendor lock-in. These concerns are recognised within Article 28 of the GDPR, which establishes obligations governing relationships between data controllers and data processors. However, compliance challenges frequently arise when institutions lack visibility into how vendors develop, train or deploy AI systems (Jin et al., 2025). Effective governance therefore requires clear contractual safeguards, ongoing oversight mechanisms and regular assessments of vendor compliance with institutional data protection standards.
Accordingly, vendor governance in higher education should be treated not merely as a technical procurement matter but as an institutional responsibility affecting student rights, educational participation and trust in university systems.
The widespread use of generative AI technologies has introduced new challenges relating to the integrity and reliability of information (Gruenhagen et al., 2024). Large language models and other content-generation systems can produce convincing but inaccurate outputs, including fabricated references, misleading academic content and erroneous factual claims (Alhitmi et al., 2024). Within higher education, such outputs may affect student learning, assessment processes, research quality and institutional decision-making.
These concerns are especially significant in universities because the reliability and originality of information are central to teaching, assessment and research. AI-generated inaccuracies may be incorporated into student assignments, teaching materials, literature reviews or research outputs, while generative AI also complicates established approaches to authorship and originality. Consequently, misinformation in higher education is not merely a general information-quality problem but may directly affect assessment validity, academic standards, research credibility and academic integrity.
The challenge extends beyond individual misuse. AI-generated misinformation can undermine confidence in educational content and complicate efforts to verify the authenticity of academic work. Moreover, institutions increasingly face difficulties in distinguishing between human-generated and AI-generated outputs. Where AI-generated outputs involve the processing of personal data, Article 5(1)(d) and (f) of the GDPR require accuracy and integrity and confidentiality, while Article 25 requires data protection by design and by default. Because generative AI tools can produce plausible but factually inaccurate content, their use also raises concerns about the integrity of student assessments and the reliability of AI-assisted research. Ensuring data accuracy, transparency and verification therefore represents a critical component of responsible AI governance and aligns closely with broader regulatory requirements relating to data integrity and accountability.
The integration of AI technologies has expanded the cybersecurity threat landscape within higher education institutions. AI-powered systems often process large volumes of sensitive information and may create new attack surfaces that can be exploited by malicious actors. Biometric systems, learning analytics platforms and cloud-based AI applications are particularly attractive targets due to the value and sensitivity of the data they contain (Binhammad et al., 2024).
Universities may face distinctive cybersecurity exposure because they operate complex and often decentralised digital environments serving large and frequently changing populations of students, academic staff, researchers, administrators and external collaborators. Their systems may simultaneously contain student records, research data, assessment materials, financial information and institutional records. Cybersecurity incidents can therefore disrupt not only administrative functions but also teaching, assessment, research activities and access to essential student services.
Recent studies have also highlighted the growing sophistication of cyberattacks supported by AI technologies, including advanced phishing campaigns, automated social engineering techniques and attacks targeting machine learning systems themselves (Nikkhah et al., 2024). Given the increasing frequency and complexity of such threats, universities must adopt proactive cybersecurity strategies that combine technical safeguards with organisational governance measures, staff awareness programmes and continuous risk monitoring. Article 32 of the GDPR requires controllers and processors to implement appropriate technical and organisational measures, including, where appropriate, pseudonymisation, encryption and regular security testing. Article 33 requires controllers to notify the competent supervisory authority without undue delay and, where feasible, within 72 hours of becoming aware of a personal data breach, unless the breach is unlikely to result in a risk to the rights and freedoms of natural persons.
The cybersecurity challenge may be compounded in the Kuwaiti context, where AI adoption is expanding faster than AI-specific regulation (Alboloushi et al., 2026). Therefore, universities there must rely largely on their own institutional capacity to manage increasingly sophisticated attack methods.
AI systems are only as reliable as the data and assumptions upon which they are built. When training data reflects historical inequalities or existing biases, AI systems may reproduce or amplify discriminatory outcomes. In higher education, algorithmic bias may influence admissions decisions, student performance predictions, scholarship allocation, or other forms of automated decision-making (Idowu et al., 2024).
The consequences of algorithmic bias are particularly significant in higher education because these decisions may directly affect access to education, academic progression and future professional opportunities. Unlike many routine automated decisions, an adverse outcome concerning admission, assessment, scholarship allocation or academic progression may have long-term consequences for a student’s educational trajectory.
Such outcomes can undermine fairness, transparency and trust in institutional processes. The challenge is particularly significant where decision-making systems operate as ‘black boxes’, limiting the ability of students, faculty members or administrators to understand how outcomes have been generated. Article 22(1) of the GDPR provides individuals with the right not to be subject to a decision based solely on automated processing, including profiling, where that decision produces legal effects concerning them or similarly significantly affects them, subject to the exceptions set out in Article 22(2). Accordingly, the GDPR and the EU AI Act provide important safeguards relating to automated decision-making, accountability and human oversight, highlighting the need for institutions to ensure that AI-assisted decisions remain subject to meaningful review. Article 14 of the EU AI Act 2024 reinforces this approach through its human oversight requirements for high-risk AI systems.
Within higher education, meaningful human oversight should therefore include mechanisms through which students can understand, question and seek review of consequential AI-assisted decisions, particularly where those decisions affect admission, assessment, progression or access to institutional opportunities.
In Kuwait, this challenge may intersect with additional equity dimensions. Gender and socioeconomic status, for example, are possible axes along which biases in training data could produce discriminatory outcomes in automated educational decision-making, a question that has yet to be examined empirically in the Kuwaiti context.
Traditional models of informed consent become increasingly difficult to operationalise in AI-driven environments because individuals often lack sufficient understanding of how complex algorithms collect, process and utilise personal information (Cohen & Slottje, 2024). AI systems frequently process large volumes of personal data through complex and often opaque algorithms, making it difficult for students and staff to fully understand how their information is collected, analysed, shared and retained. As a result, consent may be obtained formally while failing to meet the substantive objective of informed decision-making.
Scholars have argued that conventional privacy notices are frequently characterised by excessive length, technical complexity and limited accessibility, reducing the likelihood that individuals meaningfully engage with their contents (Cate & Mayer-Schönberger, 2013; Susser, 2019). These concerns are particularly relevant in higher education, where students may be required to use institutional learning platforms, assessment systems and AI-enabled services as part of their academic activities. In such circumstances, consent may not be entirely voluntary because access to essential educational services is often contingent upon acceptance of institutional data processing practices.
This creates a distinctive power asymmetry between institutions and students. A student may formally consent to the use of an AI-enabled platform yet have no realistic alternative where that platform is required for accessing course materials, submitting assessments or receiving university services. The higher education context therefore raises a particularly important distinction between formal consent and genuinely voluntary choice.
The challenge is further intensified by the increasing use of learning analytics, predictive modelling and generative AI systems that continuously collect and process personal information throughout the educational lifecycle (Gruenhagen et al., 2024). Students may therefore be unaware of the full extent of data processing activities or the secondary uses of their information. Taken together, concerns relating to privacy, transparency, algorithmic fairness and accountability may undermine student trust in AI systems, potentially limiting the effectiveness of AI adoption within higher education institutions (Nazaretsky et al., 2025).
The GDPR seeks to address these concerns through Article 4(11), which defines consent as a freely given, specific, informed and unambiguous indication of the data subject’s wishes, while Article 7 establishes the conditions for valid consent. Article 7(4) is particularly relevant in assessing whether consent is freely given where access to a service is made conditional on consent to processing that is not necessary for the provision of that service. However, compliance with formal legal requirements does not necessarily guarantee meaningful understanding by data subjects. In jurisdictions where AI governance frameworks remain underdeveloped, including Kuwait, additional challenges may arise due to varying levels of digital literacy, linguistic barriers and limited awareness of individual data rights. Consequently, institutions adopting AI technologies should move beyond reliance on traditional consent models and develop more transparent, accessible and user-centred approaches to data governance.
A final challenge concerns the broader governance structures required to oversee AI adoption (Alboloushi et al., 2026). While many institutions have implemented AI technologies, governance arrangements often remain fragmented, reactive or underdeveloped. Questions relating to responsibility, accountability, ethical oversight and regulatory compliance frequently remain unresolved.
This fragmentation can be particularly pronounced in higher education because responsibility for AI may be distributed across academic departments, information-technology units, university leadership, research offices, student services and external technology providers. Determining who is responsible for approving, monitoring and reviewing AI systems may therefore be more complex than in more centralised organisational environments.
Universities must also consider the relationship between AI governance and established academic values. AI systems used to monitor teaching activity, assess student work, detect academic misconduct or support institutional decision-making may affect academic discretion, intellectual autonomy and the professional responsibilities of faculty members. Governance in higher education therefore requires consideration not only of legal and technical compliance but also of whether AI deployment is compatible with the institution’s educational and academic mission.
This challenge is particularly evident in jurisdictions where AI-specific legislation is still emerging. In Kuwait and many other non-EU jurisdictions, institutions increasingly adopt AI technologies without comprehensive domestic regulatory frameworks governing their use. Consequently, universities must often rely on international standards, professional guidelines and internal governance policies to manage AI-related risks. The absence of clear governance structures may create uncertainty regarding accountability and weaken stakeholder confidence in institutional AI initiatives.
Collectively, these challenges demonstrate that effective AI governance requires more than technical safeguards or regulatory compliance alone. Rather, institutions must adopt integrated governance approaches that address privacy, security, accountability, transparency and ethical responsibility simultaneously. These challenges form the basis for the conceptual framework proposed in the following section.
Although the seven challenges identified in this section are not exclusive to higher education, their interaction with sensitive and longitudinal student data, compulsory use of institutional platforms, consequential educational decision-making, academic integrity obligations, academic values and distributed university governance gives them a distinctive form within the higher education environment. This sector-specific context therefore justifies the need for governance arrangements that respond to the particular institutional relationships and responsibilities of universities rather than simply transferring general AI-governance models into educational settings.
7. Conceptual Framework for Managing AI in Higher Education
Drawing upon the literature review, regulatory analysis and theoretical perspectives discussed earlier, this study proposes a conceptual framework for managing AI innovation and data security in higher education. The framework is designed to assist higher education institutions in balancing the opportunities presented by AI technologies with the governance responsibilities necessary to ensure ethical, secure and trustworthy deployment. It integrates three interdependent pillars: AI Innovation, Governance Challenges and Governance and Compliance Mechanisms. The framework is conceptual. It was developed from the literature and the regulatory analysis and has not yet been empirically validated so it should be read as a proposal to be tested rather than as a proven governance tool.
The framework adopts a dynamic and interconnected structure consisting of three mutually reinforcing pillars. The proposed framework is illustrated in Figure 1, which depicts the relationship between AI innovation, governance challenges and the governance mechanisms required to support responsible AI adoption in higher education institutions.

Pillar I: AI Innovation
The first pillar captures the principal drivers of AI adoption within higher education. These include personalised learning systems, intelligent tutoring systems, predictive analytics, automated assessment tools, generative AI applications and AI-enabled administrative systems. Collectively, these technologies support enhanced learning experiences, operational efficiency, research productivity and data-driven decision-making. However, the benefits associated with AI adoption also generate new governance responsibilities that institutions must address.
Pillar II: Governance Challenges
The second pillar identifies the principal risks and challenges associated with AI deployment, corresponding to the seven challenges examined in Section 6. These include data privacy and re-identification risks, third-party vendor dependence, misinformation and data integrity concerns, cybersecurity vulnerabilities, algorithmic bias, inadequate consent mechanisms and governance, accountability and regulatory uncertainty. These challenges represent the primary governance pressures created by AI adoption and require continuous monitoring and assessment.
Pillar III: Governance and Compliance Mechanisms
The third pillar encompasses the governance measures available to institutions for managing AI-related risks. These include compliance with applicable legal and regulatory frameworks, institutional AI policies, ethical review procedures, data governance protocols, cybersecurity controls, transparency measures, human oversight mechanisms and stakeholder engagement processes. The objective of this pillar is not merely regulatory compliance but the development of sustainable governance structures capable of supporting responsible AI adoption over time.
The three pillars are linked through continuous feedback loops. As AI technologies evolve, new risks and governance challenges emerge, requiring institutions to adapt their governance mechanisms accordingly. Effective AI governance therefore operates as an ongoing process of assessment, monitoring and institutional learning rather than a one-time compliance exercise. This iterative process is illustrated through the hypothetical application presented in Section 7.3.
The framework is grounded in three complementary theoretical perspectives. Responsible Innovation Theory informs the framework’s emphasis on anticipation, reflexivity, inclusion and responsiveness, encouraging institutions to consider the broader societal and ethical implications of AI adoption (Stilgoe et al., 2013). Information Privacy Theory provides the foundation for the framework’s focus on data protection, consent, transparency and individual autonomy (Clarke, 1999; Westin, 1967). Institutional Theory explains how governance practices become embedded within organisational structures through regulatory requirements, professional norms and stakeholder expectations (DiMaggio & Powell, 1983).
The integration of these perspectives enables the framework to address AI governance from normative, individual and organisational dimensions simultaneously. As a result, the framework moves beyond a purely technical or legal understanding of AI governance and adopts a more holistic approach centred on responsible institutional stewardship.
Framework Component | Theory | Principle Drawn On |
Pillar I: identifying AI applications and their expected benefits before adoption | Responsible innovation | Anticipation: weighing benefits against possible consequences before a technology is introduced |
Pillar II: data privacy and re-identification (6.1); inadequate consent (6.6) | Information privacy | Privacy as individual control over personal information (Westin, 1967); informational and behavioural privacy (Clarke, 1999) |
Pillar II: vendor dependence and data control (6.2); cybersecurity (6.4) | Information privacy | Loss of control over where data flow and who can access them |
Pillar II: misinformation and data integrity (6.3); algorithmic bias (6.5) | Responsible innovation | Anticipation and reflexivity: questioning how AI outputs may affect fairness, assessment and academic integrity |
Pillar II: governance, accountability and regulatory uncertainty (6.7) | Institutional | Weak coercive pressure where AI-specific law is absent; reliance on normative and mimetic pressure |
Pillar III: data minimisation, plain-language notices, routes to question AI-assisted decisions | Information privacy | Restoring practical control to students rather than relying on formal consent alone |
Pillar III: benchmarking against the GDPR and EU AI Act; institutional AI policies; peer-institution guidance | Institutional | Coercive, normative and mimetic isomorphism (DiMaggio & Powell, 1983) |
Pillar III: stakeholder engagement and ethical review | Responsible innovation | Inclusion and reflexivity (Stilgoe et al., 2013) |
Feedback loop and periodic review linking Pillar III back to Pillars I and II | Responsible innovation | Responsiveness: adjusting use of a technology as new risks appear |
Importantly, the three theoretical perspectives are not intended to correspond individually to separate pillars of the framework. Rather, they operate across the framework simultaneously. Responsible Innovation Theory informs not only the adoption of AI technologies but also the anticipation and management of governance risks and the responsiveness of institutional safeguards. Information Privacy Theory influences both the identification of privacy-related risks and the design of governance mechanisms concerning consent, transparency and data control. Institutional Theory, meanwhile, helps explain how innovation choices, governance challenges and compliance practices are shaped by regulatory pressures, professional norms and institutional expectations. The framework therefore integrates the three perspectives across its pillars rather than treating them as separate theoretical foundations for individual components.
Some examples show how the theories shaped the content of the framework. Information Privacy Theory, and in particular Westin’s (1967) view of privacy as control over one’s own information, led to re-identification risks (Section 6.1) and inadequate consent (Section 6.6) being identified as distinct challenges: both describe situations in which students lose practical control over how their data are combined and used, even where formal consent has been given. The same perspective explains why Pillar III includes data minimisation, accessible transparency notices and routes for students to question AI-assisted decisions, rather than relying on consent forms alone. Institutional Theory informed the treatment of regulatory uncertainty (Section 6.7). Because coercive pressure from AI-specific law is weak in Kuwait, governance there is expected to develop mainly through normative and mimetic pressures. Therefore, the framework includes in Pillar III the benchmarking of institutional policies against the GDPR and the EU AI Act and the use of professional and peer-institution guidance, as the Kuwait University GenAI Policy already does in part (Section 4.1). Responsible Innovation Theory shaped the process elements of the framework: anticipation corresponds to the risk assessment carried out before a system is deployed, inclusion to the participation of students and faculty in governance bodies and responsiveness to the feedback loop that carries monitoring results back to Pillars I and II.
Table 4 sets out which parts of the framework draw on which theory and which principle of that theory is used. Some components draw on more than one theory, which is consistent with the point made above that the theories operate across the pillars rather than one per pillar.
The links in Table 4 are analytical. They show how the theories shaped the design of the framework, not that the framework has been shown to work in practice. The framework has not yet been empirically validated and it should not be treated as a tested governance tool until it has been applied and evaluated in real institutional settings (see Section 8).
The framework is intended to function as both an analytical tool and a practical governance guide. From an analytical perspective, it enables researchers to examine the relationships between AI adoption, governance challenges and institutional responses. It may also assist in identifying gaps between existing governance practices and emerging regulatory expectations.
From a practical perspective, the framework provides institutional leaders and policymakers with a structured approach for evaluating AI initiatives and developing governance strategies. Universities may use the framework to map existing AI applications, identify associated risks, assess governance readiness and prioritise compliance and oversight measures. By doing so, institutions can strengthen transparency, accountability and stakeholder trust while supporting innovation.
Stage | Pillar | Main Question | Responsible Body |
1. Proposal | I | What is the system for, what benefits are expected and what data will it use? | Proposing unit (e.g., a faculty, student affairs or admissions) |
2. Risk screening | II | Does the system process personal or sensitive data, support consequential decisions or depend on an external vendor? Would it fall within a high-risk category under the EU AI Act? | Data protection officer and IT security unit |
3.Impact assessment | II | What privacy, bias, security and integrity risks arise, and which students or staff could be affected? | Data protection officer, with input from student and faculty representatives |
4. Decision | III | Should the system be approved, approved with conditions or rejected? | Institutional AI governance committee (e.g., chaired by a vice-president, with legal, IT, academic and student members) |
5. Contracting and safeguards | III | Do the vendor terms cover data processing, security, audit rights and data location? | Procurement and legal office |
6. Operation and review | I–III | Is the system working as intended, are outcomes fair across student groups, and can students request human review? | System owner in the proposing unit, reporting periodically (e.g., annually) to the governance committee |
In practice, a university could apply the framework through a review procedure that is triggered whenever a unit proposes a new AI system or a significant change to an existing one. Table 5 sets out one possible sequence of steps and the bodies that could be responsible for each decision. The allocation of responsibilities is illustrative only and the proposed framework has not been tested in an actual institutional setting. Thus, institutions would assign these roles according to their own structures and smaller institutions may combine several of them in a single committee.
For the impact assessment stage, Article 35 of the GDPR (data protection impact assessment) and Article 27 of the EU AI Act (fundamental rights impact assessment by certain deployers of high-risk systems) offer established models that a Kuwaiti institution could adapt even though it is not legally bound by either instrument.
Illustrative Scenario: An Early-Warning System at a Hypothetical Kuwaiti University
Step | Risk Identified (Pillar II) | Safeguard (Pillar III) | Responsible | Basis |
1. Data inputs | Grades, attendance and learning-platform activity are combined into detailed profiles that could identify students (6.1) | Use only the academic data needed for the prediction; exclude health and family data; pseudonymise records; set a retention period | Data protection officer | Binding: Law No. 20/2014, Arts. 32, 35 Reference: GDPR Arts. 5(1)(b)–(c), 35 |
2. Vendor | The system is hosted by an external provider, possibly outside Kuwait, which might reuse student data (6.2) | Data processing agreement covering purpose, data location, audit rights, deletion at contract end and no use of student data to train the vendor’s models | Procurement and legal office | Binding: Law No. 20/2014, Art. 35; NCSC requirements (public universities) Indirect: CITRA Res. 26/2024 if the provider is licensed Reference: GDPR Art. 28 |
3. Model testing | A model trained on past cohorts under-predicts success for some groups, e.g., by gender or type of secondary school (6.5) | Test accuracy and error rates by group before launch and each semester; report results to the committee | System owner, with the institutional research unit | Reference: AI Act Arts. 10(2)(f)–(g), 26(5) |
4. Use of flags | Risk flags become de facto decisions about students (6.5, 6.7) | Flags go only to academic advisers; no automatic change to registration, scholarships or sanctions; advisers record their decision | Dean of student affairs; academic advisers | Reference: GDPR Art. 22; AI Act Arts. 14, 26(2) |
5. Information to students | Students do not know they are being profiled, and consent is not freely given (6.6) | Plain-language notice in Arabic and English; students can see their own risk status and ask for correction | Data protection officer; student affairs | Binding: Law No. 20/2014, Arts. 32, 33, 36 Reference: GDPR Arts. 13, 15, 21; AI Act Arts. 26(11), 86 |
6. Security | Breach of, or unauthorised access to, the combined dataset (6.4) | Role-based access, encryption, logging and an incident response plan | IT security unit | Binding: Law No. 20/2014, Art. 35; NCSC requirements (public universities) Reference: GDPR Arts. 32–33 |
7. Approval and review | No clear owner for the system’s outcomes (6.7) | Committee approves with conditions; review after the first semester and then annually; suspension if conditions are not met | AI governance committee | Institutional policy |
In the absence of institutional case material, a hypothetical example is used to illustrate how the framework might be applied in practice. The scenario considers a Kuwaiti university planning to introduce an AI-based predictive analytics system to identify students at risk of academic failure. Under Pillar I, the institution would first identify the potential benefits of the system, including earlier academic intervention and improved student support. Pillar II would involve examining the risks associated with processing sensitive student data, potential bias in the algorithm, difficulties in understanding its predictions, and reliance on an external technology provider. Under Pillar III, the university would respond through appropriate data-governance controls, human oversight, transparency measures, vendor safeguards and procedures allowing consequential AI-assisted decisions to be reviewed. Following Table 5, the proposal would come from the unit responsible for student support, the data protection officer would screen it and lead the impact assessment, the AI governance committee would decide whether the system may proceed and on what conditions, and the procurement and legal office would negotiate the vendor safeguards. Rather than the system itself, academic advisers would remain responsible for any intervention with an individual student.
Table 6 follows the scenario from risk identification to the safeguards adopted and the body responsible for each. The last column distinguishes between safeguards based on legally binding requirements for Kuwaiti universities and those derived from non-binding governance references, consistent with Table 2.
The dynamic nature of the framework becomes evident where these stages interact. If monitoring reveals that the predictive system produces inaccurate or disproportionately adverse outcomes for particular student groups, this would generate a new governance concern under Pillar II and require the institution to reconsider the system’s design or use under Pillar I and strengthen its oversight and accountability mechanisms under Pillar III. The feedback process therefore allows governance measures to evolve alongside the technology and the risks identified during its operation. This example is hypothetical. It is intended to show how the framework could be used and the framework has not yet been tested in an actual institutional setting.
The framework is designed to be sufficiently flexible to operate across diverse regulatory environments. While particularly informed by the GDPR and the European Union AI Act 2024, its principles remain applicable to jurisdictions where AI-specific regulation is still emerging, including Kuwait and other countries. In such contexts, the framework may serve as a reference model for the development of institutional AI governance policies and practices.
8. Discussion
The findings of this study highlight the growing complexity of AI governance within higher education and demonstrate that effective governance requires a balance between technological innovation, data protection, ethical responsibility and institutional accountability. By integrating insights from the literature review, regulatory analysis and the proposed conceptual framework, several important observations emerge.
First, the findings reinforce the central importance of data privacy within AI governance while demonstrating that privacy-related risks extend beyond traditional concerns regarding data collection and storage. The analysis reveals that re-identification risks, inadequate consent mechanisms, third-party vendor dependence and cybersecurity vulnerabilities represent interconnected dimensions of a broader governance challenge. This finding supports Information Privacy Theory, which conceptualizes privacy not merely as the protection of information but as the preservation of individual control over how personal data is collected, processed and utilized. Consequently, governance strategies that rely solely on compliance-oriented data protection measures are unlikely to provide sufficient protection unless accompanied by broader institutional mechanisms promoting transparency, accountability and stakeholder trust.
Second, the findings demonstrate the increasingly significant influence of regulatory frameworks on institutional AI governance. The combined effect of the GDPR and the EU AI Act 2024 reflects a broader shift towards risk-based governance models that seek to balance innovation with the protection of fundamental rights. However, the analysis suggests that compliance alone is insufficient to ensure responsible AI adoption. Institutions must also develop internal governance capabilities capable of anticipating emerging risks, monitoring evolving technologies and responding to stakeholder concerns. This observation aligns with Institutional Theory, which suggests that sustainable organizational change is shaped not only by external regulatory pressures but also by the internalization of professional norms and governance expectations.
Existing research has proposed several approaches to AI governance in higher education. Chan’s (2023) AI Ecological Education Policy Framework organizes institutional AI policy around three dimensions: pedagogical, governance and operational dimensions. Similarly, Shahid et al. (2026) propose a multi-dimensional framework for generative AI in higher education, emphasizing responsible use of large language models, academic integrity, AI literacy, institutional accountability and alignment with national AI policy. More recently, Khoza et al. (2026) identify growing convergence among university AI-governance approaches around principles such as transparency, fairness and accountability, while also highlighting a movement towards more proactive and adaptable governance.
The framework proposed in this study complements these approaches but differs in its analytical structure and regulatory emphasis. Rather than organizing governance primarily around institutional functions or specific forms of generative AI, it explicitly connects three elements: the technologies and benefits driving AI adoption, the governance risks arising from that adoption, and the legal and institutional mechanisms available to address those risks. It also integrates the GDPR and EU AI Act as comparative regulatory benchmarks, thereby linking higher-education governance concerns with more specific requirements concerning data protection, risk management, transparency and human oversight.
The framework further treats these elements as interdependent rather than static categories. As illustrated in Section 7.3, the identification of new risks may require changes to both the way an AI system is deployed and the governance mechanisms surrounding it. The contribution of the framework therefore lies not in replacing existing higher-education AI-governance models, but in providing a complementary structure that connects innovation, higher-education-specific governance risks and regulatory responses through an iterative process.
The findings also highlight the growing importance of third-party vendor governance. As universities increasingly rely on external AI providers, institutional control over data processing activities becomes more fragmented and difficult to monitor. Many governance risks originate not from institutional misuse of AI but from insufficient oversight of external technology providers and the complex cross-border data ecosystems in which these providers operate. Consequently, institutions should strengthen vendor governance through robust procurement procedures, contractual safeguards, data processing agreements and ongoing monitoring mechanisms. Such measures are particularly important where AI services involve the transfer or processing of sensitive educational data across multiple jurisdictions.
The Kuwaiti context illustrates both the opportunities and challenges associated with AI adoption in environments where AI-specific regulation remains underdeveloped. While higher education institutions are increasingly embracing AI technologies to support teaching, learning, research and administration, governance structures have not always evolved at the same pace. This creates the possibility that technological adoption may outstrip institutional governance capacity, increasing exposure to privacy, security and accountability risks. At the same time, the relative absence of comprehensive domestic AI regulation creates an opportunity for institutions to adopt proactive governance practices informed by international standards and emerging best practices. From an Institutional Theory perspective, this reflects the importance of normative governance mechanisms in shaping responsible organisational behaviour in the absence of strong coercive regulatory pressures.
The findings suggest that the long-term success of AI adoption in higher education will depend not only on technological capability but also on the development of governance frameworks capable of fostering transparency, accountability, privacy protection, security and stakeholder trust. Universities should therefore move beyond reactive compliance approaches and adopt integrated governance strategies that embed ethical considerations, risk management and regulatory awareness throughout the AI lifecycle. The proposed framework contributes to this objective by providing a structured approach through which institutions can balance innovation with responsible governance and thereby support the sustainable integration of AI within higher education.
While the proposed framework provides a theoretically grounded approach to AI governance in higher education, it has not yet been empirically validated. Future research should examine its application across different institutional and regulatory contexts, particularly within emerging AI governance environments such as Kuwait.
9. Conclusion
The rapid integration of AI into higher education has created significant opportunities for enhancing teaching, learning, research and institutional administration. At the same time, it has introduced complex challenges relating to data privacy, cybersecurity, algorithmic bias, transparency, accountability and regulatory compliance. As universities increasingly adopt AI-enabled systems, the development of effective governance mechanisms has become a strategic necessity rather than a purely technical or legal concern.
This paper examined the governance implications of AI adoption in higher education through a review of the literature, an analysis of the GDPR and the European Union AI Act 2024 and the application of Responsible Innovation Theory, Information Privacy Theory and Institutional Theory. The findings identified seven interconnected categories of governance challenges: data privacy and re-identification risks, third-party vendor dependence, misinformation and data integrity concerns, cybersecurity vulnerabilities, algorithmic bias, inadequate consent mechanisms, and governance, accountability and regulatory uncertainty. These challenges demonstrate that AI governance extends beyond regulatory compliance and requires a broader institutional approach that integrates ethical, technical, organisational and legal considerations.
To address this challenge, the study proposed a three-pillar conceptual framework integrating AI innovation, governance challenges and governance and compliance mechanisms. The framework contributes to the existing literature by providing a structured approach through which institutions can evaluate AI adoption holistically, balancing innovation objectives with the requirements of privacy protection, accountability, transparency and stakeholder trust. By linking technological opportunities with governance responsibilities, the framework seeks to move beyond the traditional view that governance constrains innovation and instead positions governance as a key enabler of sustainable and trustworthy AI adoption.
The significance of this contribution extends beyond the European regulatory environment. While the framework is informed by the GDPR and the EU AI Act 2024, its underlying principles are equally relevant to jurisdictions where AI governance frameworks remain under development. In Kuwait, where higher education institutions are increasingly embracing AI technologies as part of broader digital transformation strategies, the framework offers a practical reference point for strengthening governance capacity and supporting responsible innovation. The framework remains conceptual, and its practical value has yet to be established through empirical research. Case studies in Kuwaiti universities could help assess how well it works in practice.
The long-term success of AI in higher education will depend not solely on technological advancement but on the ability of institutions to govern these technologies responsibly. Universities that embed transparency, accountability, privacy protection and ethical oversight into their AI strategies will be better positioned to realise the benefits of AI while maintaining the trust of students, staff, regulators and society. As AI continues to reshape higher education, governance should not be viewed as a response to innovation but as an essential condition for its sustainable and responsible development.
During the preparation of this work, the authors utilised an AI platform for proofreading and grammar corrections. Afterward, they reviewed and edited the content as necessary and take full responsibility for the publication’s content.
Conceptualization, M.H.A. and B.A.; methodology, M.H.A. and B.A.; validation, M.H.A. and B.A.; formal analysis, M.H.A. and B.A. writing—original draft preparation, M.H.A. and B.A.; writing—review and editing, M.H.A. and B.A.; visualization, M.H.A. and B.A.; supervision, M.H.A. and B.A. All authors have read and agreed to the published version of the manuscript.
Not applicable.
The authors declare no conflicts of interest.
