Guiding the Uncharted: The Emerging (and Missing) Policies on Generative AI in Higher Education

Meruyert ARISTOMBAYEVA¹Ryskhan SATYBALDIYEVA¹Barani Maung Maung²Dongwon Lee^3*

• ¹Satbayev University, Almaty, Kazakhstan
• ²University of Oxford, Oxford, United Kingdom
• ³The Pennsylvania State University (PSU), University Park, United States

Soon AI may move beyond its role as a mere tool to funcBon as a creaBve agent-potenBally even as a virtual student or professor-capable of generaBng original artworks and contribuBng to research leadership. However, it remains unclear whether educaBonal insBtuBons are adequately prepared for such a rapid integraBon of AI into the educaBonal and research processes. This quesBon becomes parBcularly relevant in the context of rapid advancement of large language models (LLMs) and generaBve arBficial intelligence (GenAI) and their potenBal to transform both the landscape of scienBfic research and educaBonal methodologies. This work therefore examines how educaBonal insBtuBons are responding to the integraBon of AI into research and educaBon. Specifically, we analyzed policies and guidelines regulaBng the use of GenAI in both general universiBes and art-focused insBtuBons, and conducted a strategic review of insBtuBonal approaches, along with a content analysis of some curricula related to GenAI implementaBon. Based on the analysis, we posit that current GenAI policies in higher educaBon are largely reacBve, unevenly implemented across regions and disciplines, and oWen fail to address research-specific use cases and the disBnct challenges faced by art-focused insBtuBons. This finding aligns with recent work showing that insBtuBons tend to conform to external regulatory, normaBve, and mimeBc pressures in their adopBon of GenAI, oWen prioriBzing legiBmacy and compliance over proacBve strategic vision [14]. From an educaBonal perspecBve, researcher further argue that Bloom's Taxonomy requires revision to address the cogniBve, affecBve, and metacogniBve demands of AI-assisted learning, underscoring the need for insBtuBonal policies that not only regulate GenAI but also foster criBcal thinking, ethical reasoning, and iteraBve learning processes in higher educaBon [15]. UniversiBes worldwide face the dual nature of using these technologies [5,6], navigaBng the potenBal for innovaBon alongside the need to address ethical and pracBcal concerns [2] that may disproporBonately affect students and faculty [8]. Notably, current guidelines for the use of GenAI in higher educaBon tend to represent a reacBve response, oWen embedded within modern technology programs. The discourse surrounding GenAI in higher educaBon is mulBfaceted, encompassing discussions on academic integrity, innovaBve teaching programs, and the potenBal for GenAI to enhance student outcomes [2,3]. With the release of ChatGPT, these issues have gained a parBcular relevance. While many universiBes have experienced challenges associated with plagiarism, data confidenBality and privacy [1,2,10], art-focused universiBes have encountered issues related to copyright protecBon and assessment of the originality of creaBvity [4].In this paper, we comprehensively reviewed publicly available policies and guidelines regarding the use of GenAI in educaBon and research from higher educaBon insBtuBons in the US, Europe and Central Asia. General universiBes from the US and Europe were selected from among the leaders of the QS World University Rankings 2024. For the US, we focused on the top 20 universiBes according to the QS World University Rankings 2024 (general category). The list of European universiBes was compiled with explicit consideraBon for geographical diversity, ensuring representaBon from different European countries. StarBng from the top of the QS rankings, we applied an addiBonal rule: if several consecuBve insBtuBons belonged to the same country (e.g., mulBple from the UK), we selected the highest-ranked one and then moved to the next university from another European country. This procedure helped avoid overrepresentaBon of a single country while sBll capturing leading insBtuBons. If a university appeared in the ranking but did not provide publicly available GenAI-related policies or guidelines (i.e., official documents explicitly outlining recommendaBons or regulaBons), it was excluded from the sample. In such cases, we conBnued down the QS list unBl a suitable insBtuBon was idenBfied. References to projects, pilot iniBaBves, or informal menBons of GenAI were not considered sufficient for inclusion. It is important to note that although many insBtuBons arBculate strategies (e.g., vision statements, curricular integraBon plans), these do not always translate into concrete policies (e.g., official guidelines for faculty and students). Throughout the analysis we therefore treat strategies as indicaBve of insBtuBonal intent, while policies are considered evidence of operaBonalized regulaBon.The list of art universiBes was compiled based on the QS World University Rankings by Subject (Art & Design), but not many relevant policies were publicly available on the universiBes' websites. We searched for exisBng policies and guidelines using the following keywords on Google: "GeneraBve AI Policy in art universiBes USA", "GeneraBve AI Policy in art universiBes Europe [countries]", "GeneraBve AI Policy in higher educaBon".UniversiBes in Central Asia were selected based on the QS World University Rankings by Region: Central Asia 2024, with consideraBon given to insBtuBons from all five Central Asian countries. Since many universiBes in the region from the list did not have policies or guidelines available on their official websites, we conducted addiBonal searches using Google. In our search, we used keywords in both English and Russian, such as "Policy on the use of generaBve AI in Kazakhstan", "Satbayev University GeneraBve AI Guide", as well as similar formulaBons for each university and country in the region. To idenBfy art universiBes in Central Asia, we conducted a separate search using keywords such as "Art UniversiBes in Kazakhstan" and equivalent terms for other countries in the region.Figure 1 illustrates the selecBon of policies and guidelines from general and art universiBes in the US, Europe and Central Asia for this study. The analyzed documents provide informaBon on exisBng policies and guidelines for students, teaching staff, and researchers regarding the safe and ethical use of generaBve AI in teaching and research. The analysis demonstrates that many general universiBes in the USA and Europe have developed their own policies and guidelines; however, policies regulaBng the use of generaBve AI are less developed in art-focused universiBes. Policies addressing students and faculty are more prevalent than those for researchers, however, 40% of the analyzed universiBes have separate guidelines for researchers with detailed recommendaBons. In Central Asia universiBes, there is a notable absence of policies and guidelines regulaBng the use of GenAI. Only a few general universiBes, including Nazarbayev University in Kazakhstan and Westminster InternaBonal University in Uzbekistan, have published relevant documents. No policies or recommendaBons regarding the use of GenAI are found on the official websites of art universiBes in the region.To assess insBtuBonal readiness and policies regarding GenAI across different regions, we applied the theoreBcal framework proposed by Lim et al. [11], which encompasses seven dimensions of strategic planning. Building on this framework, we constructed a structured analyBcal matrix covering each dimension (e.g., vision and narraBve, curriculum integraBon, technology-centric support, human-centric support, stakeholder engagement), and mapped the content of insBtuBonal policies to these categories. Three researchers independently described the policies of each university across the seven dimensions of the Lim et al. framework. Their individual descripBons were then compared and consolidated into a unified account through discussion, which helped to ensure consistency of interpretaBon. While this approach provided a structured basis for cross-regional comparison, we acknowledge certain limitaBons: in several cases, insBtuBons did not provide comprehensive or publicly accessible documentaBon, and therefore only parBal evidence could be included, which may underrepresent the actual extent of insBtuBonal readiness. Separately, a raBng system inspired by the methodology of [7] was developed to evaluate the leniency of university policies. Table 1 presents the universiBes from US and Europe included in the analysis, along with their leniency raBngs based on a 5-point Likert scale. UniversiBes from Central Asia are excluded as most did not have GenAI policies yet. Three authors independently evaluated the policies aWer agreeing on raBng criteria. Final raBngs were determined by majority voBng. Nevertheless, we acknowledge the potenBal influence of subjecBve factors on the assessment results, which is primarily due to implicit ambiguiBes in the policy formulaBons.Most universiBes showed a moderate degree of leniency (raBngs of '3' or '4') to the integraBon of GenAI into the educaBonal process and scienBfic research, while all educaBonal insBtuBons recognize the importance of using new technologies and the need to adapt to modern realiBes. For instance, The University of Chicago's policy received a '3' for educaBon because the university allows GenAI in selected tasks, from brainstorming ideas to clarifying complex concepts, but also sets limits on its use in some academic work. The University of Edinburgh was rated '4' as they permit the use of GenAI flexibly provided it is disclosed. Only two universiBes received a raBng of '2' for educaBon: the University of Amsterdam and Ludwig Maximilian University of Munich, because they allow only limited use for narrowly defined purposes.Takeaway: These iniBal observaBons highlight a fundamental gap: while many insBtuBons recognize the transformaBve potenBal of GenAI, policy responses remain largely reacBve and fragmented, lacking a unified framework to support both educaBonal and research pracBces.Table 1. Leniency raDngs on a 5-pt Likert scale for universiDes in USA and Europe, where "1" = "Extremely restricDve" (restricts almost all types of GeneraDve AI use), "2" = "Somewhat restricDve" (allows the limited use of GeneraDve AI for narrowly defined purposes, such as grammar correcDons, text simplificaDon, or non-sensiDve data analysis. Transparency and consultaDon are required before adopDng AI tools), "3" = "Neither lenient nor restricDve" (balanced with restricDons and allowances, permiTng selected tasks such as ediDng, summarizing, and idea generaDon under clear guidelines that prioriDze transparency, data protecDon, and accuracy. However, tasks such as examinaDons, discussion-based assessments, class tests, laboratories, and pracDcals are prohibited), "4" = "Somewhat lenient" (permits most applicaDons of GeneraDve AI in academic, research, and creaDve contexts, provided ethical guidelines are adhered to and usage is disclosed. Supports tasks such as academic content creaDon, data analysis, and creaDve exploraDon, while maintaining awareness of copyright and intellectual property concerns), "5" = "Extremely lenient" (allows the use of all types of GeneraDve AI). UniversiDes with no policies are marked as '-'. University policies were iniDally evaluated between December 9-18, 2024, and subsequently reviewed and revised between April 8-14, 2025. We analyzed a total of 30 general universiBes from the United States and Europe. However, selecBng art-focused universiBes based on rankings has been challenging due to the absence of publicly available policies and guidelines. As a result of an extensive review of exisBng policies and guidelines available online, at the end, we selected 6 US and 5 European art-focused universiBes, as shown in Table 1.As previously noted, the approaches of general universiBes and art-related insBtuBons are very similar, but there are slight differences in the universiBes' strategies. Table 2 shows the key differences and common strategies for GenAI between general and art-focused universiBes.That is, in general universiBes, the focus is placed on academic ethics and the educaBonal process, whereas in art-focused insBtuBons, the focus shiWs to authorship, creaBvity and cultural implicaBons of using GenAI. We further analyzed how general universiBes and art-related insBtuBons implement GenAI across insBtuBonal pracBces by applying the framework proposed by Lim et al. [11], which idenBfies seven core areas: vision and policy, curriculum, infrastructure and resources, professional development, student support, partnerships, and research. General universiBes such as MIT, Harvard, Stanford, Oxford, and KU Leuven have adopted a structured and mulBdimensional approach to GenAI integraBon, aligning closely with the seven areas outlined in Lim et al.'s (2019) framework. These insBtuBons typically begin with strong insBtuBonal visions that emphasize transparency, academic integrity, and ethical responsibility. Faculty are supported through detailed syllabus templates, training workshops, and ethical guidelines for GenAI use in both teaching and research. On the infrastructure side, many universiBes provide secure, licensed environments (e.g., PhoenixAI in UChicago, U-M GPT in UMich, AI Sandbox in Harvard) and riskassessment protocols to ensure data privacy. Human-centric support includes consultaBons, seminars, and instrucBonal design assistance to foster criBcal engagement with GenAI. Furthermore, policy development is collaboraBve and interdisciplinary, involving IT services, legal counsel, teaching centers, and ethics boards. This comprehensive model reflects a proacBve insBtuBonalizaBon of GenAI tools into academic, administraBve, and research processes.In contrast, art-related insBtuBons such as RISD, Ringling, MICA, Prat, and CCA are sBll in the early stages of GenAI policy development. While some of them have begun experimenBng with GenAI integraBon, their efforts remain largely decentralized and course-specific. Faculty are oWen granted autonomy to set GenAI-related rules, and insBtuBonal visions emphasize creaBvity, authorship, and reflecBve use of AI rather than systemic implementaBon. Curricular integraBon is typically limited to design and wriBng assignments, with an emphasis on process, iteraBon, and ethical exploraBon. Infrastructure is oWen limited to recommended external tools (e.g., Midjourney, ChatGPT, DALL-E), with safety and atribuBon guidelines provided through libraries and teaching centers. Human-centric support includes workshops, collaboraBve assignments, and criBcal dialogue around copyright, bias, and authenBcity. While some insBtuBons (e.g., RISD, Prat) are beginning to formalize policies through provost offices or AI task forces, engagement with stakeholders is typically informal and driven by teaching units, librarians, and design labs. This botom-up and discipline-driven model aligns only parBally with Lim et al.'s framework-primarily in the areas of curriculum, vision, and student support-highlighBng an ongoing transiBon toward more formalized insBtuBonal strategies.We analyzed the curricula of those universiBes to examine how they adapt to modern technological advancements. In recent years, they have been acBvely modernizing their educaBonal programs in response to the development of AI. In parBcular, the integraBon of specialized courses on GenAI highlights its growing role in creaBve work. Many art universiBes are integraBng courses on GenAI, AI-driven wriBng, and AI techniques in art in their curricula. In May 2024, for instance, Ringling College of Art and Design in the US announced the launch of a new AI undergraduate certificate program to equip students with the knowledge and skills needed to understand and use AI in creaBve industries of all kinds. In April 2024, in addiBon, Rhode Island School of Design in the US offered new courses on GenAI such as Designing with Emerging Technologies: GeneraBve AI and Text Transformed: WriBng in the Age of AI. However, many schools have no GenAI related policies for their students or faculty.In addiBon, Figures 2 and3 show the Bmeline starBng from the emergence of GPT-3.5, when policies for the use of GenAI began to be developed in the first a few general universiBes. The Bmeline figures were compiled based on the deadlines provided in the guidelines published on the official websites of the universiBes. In cases where date informaBon was not available, the date was esBmated using the Wayback Machine service, based on the Bme of the last update of the relevant material on the site. In general, universiBes first introduced GenAI policies for educaBonal acBviBes, and only a few subsequently extended these to research acBviBes. By and large, the primary focus has been on guiding students' learning and faculty's teaching pracBces involving GenAI.Takeaway: This contrast reinforces the core argument that while general universiBes are making structured efforts toward policy development, art-focused insBtuBons are lagging, exposing a disciplinary divide that leaves creaBve fields without coherent or comprehensive GenAI guidance. Figure 3 presents a Bmeline of policy publicaBons across regions. The US and Europe have been acBvely implemenBng guidelines and policies for the use of GenAI since 2023. In general, approaches to regulaBng this process show litle to no variaBon based on regional characterisBcs. Both areas place significant emphasis on academic integrity, transparency, confidenBal data protecBon and copyright compliance. Most universiBes were rated '3' or '4'. Notably, GenAI policies for research only began to appear in 2024, almost a year aWer the iniBal guidelines for educaBon were introduced.However, in Central Asia, the process of developing and adapBng either policy is sBll in its infancy. One of the primary challenges lies in limited technological access, insufficient resources, and inadequate insBtuBonal infrastructure [12], alongside the absence of centralized naBonal policies specifically addressing the use of GenAI in educaBon. Although many Central Asian universiBes have not yet developed formal guidelines, they are taking steps to integrate these technologies into the educaBonal process and research. In parBcular, universiBes organize various seminars, round tables, and discussions, while also acBvely integraBng GenAI courses into their curricula.Takeaway: The regional disparity, especially the policy vacuum in Central Asia, underscores the global unevenness in insBtuBonal readiness, highlighBng the absence of centralized naBonal strategies and the pressing need for inclusive, region-sensiBve GenAI governance in higher educaBon. Based on the analysis, the three groups of higher educaBon insBtuBons lack clear policies or regulatory guidance on the use of GenAI. The first group includes universiBes, both general and art-focused, that do not have policies or guidelines defining the principles for the use of GenAI in research-e.g., when or whether graduate students can use GenAI to create syntheBc research data, or faculty members can use GenAI to assist in wriBng scholarly arBcles. Out of 41 universiBes, only 17 (42%) have established guidelines in research. The second group consists of art-focused universiBes around the world, only a small number of which have developed relevant guidelines or policies. Even aWer an extensive search, we have idenBfied only 6 GenAI policies from art universiBes in the USA and 5 in Europe. The third group is represented by universiBes in Central Asia countries. Out of 30 universiBes that we have examined, only 2 (6.6%) universiBes have guidelines for the use of GenAI. Based on the idenBfied characterisBcs, we have formulated the following recommendaBons for each of the three groups of educaBonal insBtuBons.Our analysis further demonstrated the value of applying Lim et al.'s [11] framework to systemaBze insBtuBonal policies. In leading U.S. and European universiBes, declaraBve statements regarding GenAI (e.g., visions of responsible and ethical use) are at least parBally reinforced by concrete measures in other dimensions, such as curriculum guidelines, provision of secure AI tools, and faculty support iniBaBves. By contrast, in art-focused insBtuBons, the imbalance between vision and pracBce is more pronounced. While creaBvity, authorship, and ethical reflecBon are consistently highlighted at the conceptual level (e.g., Ringling, RISD, MICA, Prat, RCA), these principles are only weakly supported by infrastructural investment, systemaBc curricular integraBon, or insBtuBonalized faculty and student support. Stakeholder engagement also tends to remain fragmented and project-based. Taken together, this comparison underscores that although top-ranked general universiBes oWen begin to translate their declaraBve visions into tangible pracBces, art-focused insBtuBons sBll face significant gaps between arBculated values and operaBonal mechanisms.For the first group, insBtuBons should encourage researchers to disclose the use of GenAI in research acBviBes including the wriBng and reviewing of scienBfic publicaBons and grant proposals. Such transparency is essenBal for ensuring the reproducibility and integrity of research. Another criBcal area is the standardizaBon of informed consent procedures, parBcularly in cases where parBcipant data may be used to train AI models. UniversiBes should also establish internal review mechanisms to assess potenBal risks associated with the use of GenAI in research, including the evaluaBon of algorithmic biases, model limitaBons, and their possible effects on research outcomes. AddiBonally, universiBes are encouraged to promote AI literacy among both researchers and students. This can be achieved through the development of targeted courses, workshops on AI ethics, and the integraBon of AI competency frameworks into exisBng curricula.[9] presents a comprehensive sequence of recommendaBons for integraBng GenAI into higher educaBon. These include educaBonal workshops, curriculum integraBon, ethical policy development, transparent communicaBon, feedback-driven refinement, faculty support, empirical evaluaBon, the promoBon of criBcal thinking, and ensuring equitable access. While many general universiBes have adopted these measures to varying degrees, art-focused universiBes sBll lack a coherent and unified strategy for incorporaBng GenAI technologies-highlighBng the need for clearer guidelines and policy development.For art universiBes (the second group), unique challenges arise due to the centrality of creaBve expression, authorship, and arBsBc labor. Policies for these insBtuBons must account for concerns such as job displacement, privacy and data protecBon, ethical use, social inequality, and loss of human agency, as emphasized in [9]. Moreover, parBcular atenBon should be given to preserving the authenBcity and value of creaBve labor in the age of GenAI.Finally, for the third group-universiBes in Central Asia-it is recommended that insBtuBons begin developing GenAI-related policies and guidelines by drawing on the best pracBces of leading U.S. and European universiBes, while at the same Bme adapBng them to the cultural, linguisBc, and insBtuBonal specificiBes of the region, as well as to the concrete educaBonal needs of both faculty and students. Such efforts should remain aligned with emerging global frameworks and ethical standards arBculated by internaBonal organizaBons such as UNESCO, and the OECD, thereby enabling local iniBaBves to contribute to broader global academic interoperability. As [13] emphasizes, although UNESCO's Guidance provides a suitable starBng point for policy development, local contextualizaBon and evaluaBon are essenBal for its meaningful applicaBon. In this regard, Table 2 outlines a comparaBve framework that highlights the disBncBve emphases of general and art-focused universiBes while also idenBfying their shared strategies, which can guide the formulaBon of policies that are both context-sensiBve and globally coherent.At the same Bme, the findings of this analysis should be interpreted in light of several limitaBons. First, the study covers only publicly available documents, which may lead to an underesBmaBon of the actual level of insBtuBonal readiness, while simultaneously poinBng to the limited transparency of certain universiBes. Second, the analysis was qualitaBve in nature and relied on textual interpretaBon; although the descripBons were carried out independently by mulBple researchers and subsequently reconciled collecBvely, the possibility of subjecBve