Abstract
Generative artificial intelligence has forced universities to reconsider academic integrity, assessment, teaching roles, institutional data governance and the infrastructure through which learning takes place. This study compares the governance model developed at Fudan University with emerging practices at Hungarian universities. It asks how institutions translate general principles of responsible AI into rules, organizational responsibilities, technological systems, assessment practices and capacity-building programmes. The analysis draws on Fudan's 2026 guidance for generative AI in education, supporting documents on its AI-enabled teaching infrastructure, interviews conducted with university representatives, and public policies and guidance issued by Hungarian institutions. The comparison identifies two distinct institutional patterns. Fudan combines a university-wide governance framework, a dedicated innovation centre, an integrated platform and a staged model of AI competence development. Hungarian universities have created increasingly sophisticated rules, yet these initiatives remain distributed across institutions and faculties and are strongly shaped by the European Union's rights-based legal environment. The study argues that Fudan's main relevance for Hungary lies in its organizational integration: it links regulation, pedagogy, infrastructure and professional development. Hungarian universities can adapt this logic by establishing institution-wide coordination, process-based assessment, secure AI services, transparent declaration requirements and differentiated training, while retaining European data-protection and fundamental-rights safeguards.
Keywords: artificial intelligence; higher education; AI governance; Fudan University; Hungary; academic integrity; comparative institutional analysis
1. Introduction
Generative AI has entered higher education through everyday practice rather than through a planned institutional transition. Students use language models to explain concepts, translate texts, generate code, summarize sources and draft assignments. Teachers use the same systems to prepare classes, create examples, design assessments and support research. Universities must now determine which forms of assistance strengthen learning, which forms substitute for the intellectual work being assessed, what information users may share with external platforms and how institutions can verify academic achievement.
This change affects the university as an organization. A rule on plagiarism cannot by itself govern AI-supported learning. Institutions must connect academic integrity with curriculum design, staff development, information security, procurement, data protection and quality assurance. They must also decide whether AI will remain a collection of external consumer services or become part of the university's own educational infrastructure.
China and the European Union provide two important contexts for observing this institutional transition. China's national policy treats university AI development as a strategic component of research, talent formation and economic modernization. The Ministry of Education's Action Plan for Artificial Intelligence Innovation in Higher Education assigned universities a central role in basic research, interdisciplinary training and technological application. Fudan University has translated this strategic orientation into an institution-wide model that joins governance with an AI-enabled teaching ecosystem.1
Hungarian universities operate inside the European Union's legal order. The General Data Protection Regulation establishes principles for lawful, transparent and purpose-limited processing of personal data. The EU AI Act adds a risk-based framework, identifies particular educational uses as potentially high-risk, and requires providers and deployers to promote AI literacy among the staff who use AI systems. Hungarian institutions have also issued their own rules and recommendations. ELTE, BME, the University of Miskolc, Pázmány Péter Catholic University, Semmelweis University and other institutions now address acceptable use, disclosure, data protection and assessment in different combinations.2
This study asks: How do Fudan University and Hungarian universities translate general principles of responsible AI into institutional rules, technological infrastructure, assessment practices and capacity-building programmes? It examines governance as an operational system. The comparison therefore focuses on seven dimensions: institutional responsibility; permitted and prohibited use; transparency; assessment; sanctions; competence development; and technological infrastructure.
The comparison does not present China or Hungary as uniform systems. Fudan represents one advanced Chinese institutional model. Hungarian cases demonstrate a range of responses within one national and European legal setting. The purpose is to identify how different arrangements connect principles with practice and what Hungarian institutions can learn from Fudan while preserving their own legal and academic foundations.
2. Methodology and Case Selection
The study uses a structured qualitative comparison. The Fudan case is reconstructed from the university's Guidelines for the Application of Generative Artificial Intelligence in Education and Teaching, Version 1.0, supporting material on its AI teaching assistant and technological ecosystem, and interviews with university representatives responsible for international relations and AI-related educational governance. These materials show both formal rules and the organizational mechanisms intended to implement them.3
The Hungarian side draws on institution-level and faculty-level policies and recommendations. The selected cases represent distinct governance emphases. ELTE and BME provide detailed guidance for students and teachers. The University of Miskolc has adopted a university regulation with explicit provisions on responsibility, disclosure and protected information. Pázmány Péter Catholic University connects practical recommendations with a human-centred and ethical understanding of university work. Semmelweis University approaches AI through a documentation and compliance framework suited to a health-science institution. The University of Debrecen, Széchenyi István University and other institutions illustrate competence-building, industrial collaboration and strategy formation.
The analysis codes the documents along the same seven dimensions. It distinguishes formal requirements from recommendations, institution-wide systems from faculty initiatives, and technological provisions from purely normative statements. Public documents do not reveal every implementation practice. The study therefore makes claims about documented governance and identifies inferences where institutional practice cannot be observed directly.
The study also has a temporal limitation. University AI policies change rapidly. The analysis describes the frameworks available by August 2026. The comparison should be treated as a snapshot of an evolving institutional field.
3. Two Governance Contexts
3.1 China: coordinated development and institutional experimentation
China's higher-education AI policy links scientific capacity, talent development and application. The 2018 Ministry of Education action plan instructed universities to strengthen AI research, build interdisciplinary programmes, develop teaching resources and cooperate with industry. This framework gives universities a developmental mandate. It encourages them to produce knowledge and infrastructure as well as rules.4
Fudan's approach reflects this mandate. The university treats AI as a new layer of educational infrastructure. Its governance model covers teaching, learning, assessment, professional development and platform design. The university's institutional response therefore begins with a strategic question: how can Fudan redesign educational processes so that teachers and students can use AI under human responsibility?
The model also sits within a wider Chinese governance discourse that places development and security in the same framework. National statements emphasize innovation, risk control, social benefit and participation in global governance. Fudan's Center for Global AI Innovative Governance embodies this combination at the research-policy level, while the university's educational guidance applies a comparable logic to teaching practice.5
This setting supports coordinated experimentation. A university can establish a dedicated centre, build internal platforms, provide shared tools and revise guidance as experience accumulates. Regulation functions as one component of institutional development.
3.2 Hungary and the European Union: rights, accountability and distributed adaptation
Hungarian universities respond within a legal environment that assigns specific responsibilities to data controllers, system providers and deployers. The GDPR requires a lawful basis for personal-data processing and applies the principles of purpose limitation, data minimization, accuracy, security and transparency. These duties become relevant whenever teachers or students enter personal, confidential or research data into an AI service.6
The EU AI Act adds obligations according to the purpose and risk of a system. Educational AI is not automatically high-risk. Systems that determine access, admission, assignment to educational institutions, evaluation of learning outcomes, or the appropriate level of education may fall within the high-risk category when the statutory conditions apply. Ordinary writing assistance or brainstorming usually occupies a different legal position. Universities must therefore classify concrete uses instead of treating every AI tool as one category.7
The AI Act also makes AI literacy an organizational responsibility. Providers and deployers must take measures to ensure a sufficient level of AI literacy among staff and other persons acting on their behalf, taking account of their knowledge, experience and context. For universities, this connects legal compliance with staff development.8
Hungarian institutions have translated this environment through distributed initiatives. Some universities issue central regulations; others allow faculties to define subject-specific practices. This distribution supports disciplinary autonomy. It also produces different rules across courses and uneven access to secure tools, training and implementation support.
The comparison thus begins with a structural difference. Fudan has assembled a coordinated institutional ecosystem. Hungary has developed a plural set of policies inside a strong external legal framework. The decisive question is how each arrangement guides daily academic work.
4. Comparative Analysis
4.1 Institutional governance and responsibility
Fudan connects strategic leadership with a dedicated implementation structure. Its Artificial Intelligence Education and Teaching Innovation Center coordinates guidance, training, platform development, educational research and the collection of practical experience. This organizational location matters. Teachers do not receive rules from one office, tools from another and training from a third without a common framework. The centre can align these functions and revise the system as educational use develops.9
The Fudan model distributes responsibility within this coordinated structure. University-level guidance defines common principles. Teachers specify acceptable AI use for courses and assignments. Students remain responsible for submitted work and must follow disclosure requirements. Technical teams maintain approved environments and supporting services. This division retains teacher autonomy while giving it institutional support.
Hungarian governance is more heterogeneous. ELTE has combined university-level action with faculty guidance. Its Faculty of Economics permits supportive uses and states that AI cannot replace students' creativity, decision-making and analytical work. BME faculties define practical recommendations suited to engineering and economic education. The BME Faculty of Electrical Engineering and Informatics explicitly supports the development of skilled AI users while requiring students to understand the technology's limitations. The University of Miskolc has moved toward central regulation, defining obligations for members of the university community and connecting AI use with data protection and information security.101112
These examples show substantial institutional learning. They also reveal a coordination problem. A faculty may create detailed rules without having access to an institutional AI platform or a permanent methodological unit. Teachers may carry responsibility for assessment redesign without receiving time, expertise or technical support. Hungarian universities often locate governance in documents; Fudan locates it in a document-plus-organization-plus-platform system.
The practical lesson concerns ownership. A university needs an identifiable body with authority to coordinate academic, legal, technical and pedagogical work. This body does not need to centralize every course decision. It needs to maintain common standards, secure services, templates, training and a feedback mechanism.
4.2 Permitted and prohibited use
Fudan avoids a single institution-wide list that labels every use either permitted or forbidden. Its framework combines baseline prohibitions with context-specific teacher decisions. Users may employ AI for explanation, ideation, language support, practice and other learning activities when these uses serve the educational objective. Teachers can require, permit, limit or prohibit AI for a particular task. Students must retain intellectual control and comply with the conditions of the assignment.
The baseline limits protect academic integrity, confidentiality and human responsibility. Users may not present generated content as independent work when the task requires independent authorship. They may not upload sensitive, confidential or protected material to unapproved external services. They may not use AI to fabricate data, references or research evidence. They remain responsible for factual accuracy, attribution and the final product.13
Hungarian guidance increasingly follows a similar functional logic. ELTE GTK treats AI as an assistant that can support learning, research planning and project work while preserving independent student performance. BME VIK encourages competent use and leaves teachers space to define requirements for specific subjects and tasks. The University of Miskolc distinguishes acceptable support from conduct that misrepresents authorship or breaches data and security rules. PPKE's recommendations emphasize personal responsibility, critical judgment and the preservation of human agency in academic work.1415
The main difference lies in operational support. A contextual rule asks teachers to design precise task-level instructions. Fudan supplies an institutional framework and examples for this work. In Hungary, the quality of assignment-level rules often depends on individual initiative. A general statement that AI is permitted “when the teacher allows it” transfers complexity to the teacher. The institution must provide model clauses, task categories and examples if contextual governance is to remain consistent.
A useful classification contains four levels: AI required as part of the learning objective; AI permitted with disclosure; AI permitted only for specified supporting activities; and AI prohibited because the task measures unaided performance. Such a scale gives teachers a common language and gives students predictable expectations.
4.3 Transparency and declaration
Fudan turns transparency into a procedure. Students who use generative AI must identify the tool and describe the role it played in the work. Depending on the task, a declaration can include the service, version, purpose, relevant prompts, generated material and the manner in which the student verified and revised the output. Disclosure is not treated merely as a citation question. It provides evidence about the learning process.16
This procedure serves three functions. It allows the teacher to evaluate the student's contribution. It encourages the student to reflect on how the system influenced the work. It also creates material for institutional learning: declarations reveal common practices, recurring risks and training needs.
Hungarian policies also emphasize disclosure. BME guidance explains how students should indicate AI-generated or AI-assisted content. ELTE and the University of Miskolc require transparency where AI contributes materially to submitted work. These provisions establish an important norm: concealed substitution breaches academic integrity, while declared assistance can form part of legitimate academic practice.
Implementation remains uneven because no shared declaration format exists across Hungarian higher education. Students encounter different expectations between institutions, faculties and courses. A simple institutional template would improve consistency. It should record the tool, purpose, extent of use, verification steps and the student's original contribution. High-stakes work may require an appendix with selected prompts and outputs; routine tasks may need one short statement.
Transparency must remain proportionate. Universities should avoid collecting complete conversational histories when a short declaration is sufficient. Prompt records can contain personal or confidential information. The declaration system must therefore follow data-minimization principles and define retention, access and purpose.
4.4 Assessment and evidence of learning
Generative AI weakens the evidentiary value of a polished final text. A student can produce fluent prose without understanding the argument. Universities must therefore assess learning through multiple forms of evidence.
Fudan's model treats AI use itself as an assessable competence. Teachers can evaluate whether the student framed the problem well, selected suitable tools, designed effective prompts, verified results, recognized limitations and integrated AI output into an original argument. This shifts attention from detecting machine-produced sentences to examining human judgment.17
The model supports process evidence: drafts, reflection notes, prompt excerpts, oral defence, in-class performance, peer discussion and iterative revision. Different disciplines can combine these forms differently. A programming course may assess code explanation and debugging decisions. A humanities course may assess source verification and interpretive choices. A research-methods course may require a documented audit trail.
Hungarian guidance has begun to move in the same direction. BME and ELTE materials encourage teachers to redesign tasks and make permitted use explicit. The University of Miskolc allows teaching contexts to determine the role of AI while holding the student responsible for demonstrable knowledge. These approaches create the basis for a process-oriented model.
The transition has significant workload consequences. Oral defence, staged submissions and individualized feedback require staff time. Universities need supporting technology and shared assessment design. They also need to protect students from arbitrary suspicion. AI detectors cannot reliably establish authorship and should not function as sole evidence in disciplinary decisions. A fair procedure should rely on the work process, the student's explanation, course rules and corroborating evidence.
Assessment redesign should begin with the learning outcome. If a course assesses unaided recall or performance, controlled conditions may be appropriate. If it assesses professional problem-solving, students may need to demonstrate competent AI use. If it assesses research judgment, the task should reveal how the student selects sources, tests claims and handles uncertainty. Governance becomes credible when assessment and declared educational purpose align.
4.5 Sanctions and educational correction
Fudan uses differentiated consequences. Minor failures of disclosure can lead to correction and resubmission. Misrepresentation, fabricated evidence, serious data breaches or repeated misconduct can trigger formal academic sanctions. This distinction recognizes that students are learning a new practice and that not every error has the same academic significance.18
Hungarian universities already possess disciplinary and academic-integrity procedures. AI rules must connect new forms of misconduct to these established systems. The institution should define the difference between inadequate citation, unauthorized assistance, false authorship, fabrication and security violations. Clear categories improve proportionality and due process.
An educational response is especially important during transition. Students need examples of acceptable declarations and opportunities to correct procedural mistakes. Serious deception still requires formal consequences. A transparent scale helps students understand that the institution protects both learning and fairness.
4.6 AI literacy and professional development
Fudan organizes competence development through three stages: acquiring AI knowledge, applying AI in educational practice and advancing AI through experimentation and interdisciplinary innovation. The model addresses both students and teachers. Basic literacy includes system capabilities, limitations, ethics and security. Applied competence connects tools with concrete disciplinary tasks. Advanced competence supports the design of new methods, agents, resources and “AI for X” applications.19
This progression solves a common training problem. A single general webinar cannot prepare every user. A literature teacher, engineer, administrator and data scientist need different forms of competence. Fudan's layered model provides a common foundation and permits role-specific development.
Hungarian universities possess strong but dispersed initiatives. BME builds AI competence into engineering education and provides practical guidance. ELTE combines policy with teaching resources. Debrecen offers specialized AI programmes and industry-linked development. Széchenyi István University has created an AI and Future Strategies Centre. The University of Miskolc offers applied programmes and has adopted institution-wide rules. PPKE contributes a strong human-centred ethical and philosophical perspective.
The EU AI Act reinforces the need to connect these initiatives with organizational responsibility. AI literacy should become a continuing programme rather than an optional event. A university can define a common baseline for all staff, specialized modules for teachers and administrators, and advanced laboratories for innovators. Completion records, communities of practice and teaching consultations can make training part of institutional quality assurance.
Teacher development deserves priority. Students adopt tools quickly; teachers must redesign learning environments while preserving academic standards. They need tested assignments, disciplinary examples, secure experimentation spaces and recognition for the time invested in redesign. Fudan institutionalizes this support through its innovation centre. Hungarian universities can create similar functions through a central unit connected to faculty-level experts.
4.7 Technological infrastructure and data governance
Fudan's governance model rests on an integrated technological environment. Its educational AI platform provides access to guidance, approved resources, teaching tools, communities of practice and AI-supported services. Supporting materials describe teacher assistants, course-quality analysis and connections to university computing capacity. The platform reduces dependence on uncontrolled consumer tools and gives the institution greater influence over data, access and system design.20
Technological sovereignty in this context means practical institutional control. The university can decide which tools users access, what data enter the system, how services are authenticated and how educational applications connect with existing platforms. It can also evaluate use and improve support.
Hungarian universities often rely on external services chosen by individual users. This creates unequal access and unclear data flows. Free consumer accounts may use prompts under terms that do not meet institutional requirements. Teachers and students may enter unpublished research, personal data, assessment material or partner information without understanding where it is processed.
European law makes data governance central. Universities should provide approved services with contractual safeguards, access controls and clear instructions. They should classify data and prohibit the entry of protected categories into unapproved systems. Privacy impact assessment may be necessary where a system processes personal data at scale or creates significant risks. High-risk educational systems require additional governance under the AI Act.21
An institutional platform does not require one monolithic model. A university can provide a gateway to several services, combine cloud and local models, and define access according to purpose. The essential features are identity management, logging proportionate to risk, data classification, procurement review, transparent user information and a support structure. Fudan demonstrates the organizational value of placing these features inside one educational ecosystem.
5. Transferability Across Institutional Cultures
Fudan's model cannot be copied as a complete package. Chinese and Hungarian universities operate under different laws, governance traditions, funding structures and expectations concerning institutional and individual autonomy. Transfer therefore requires functional adaptation.
5.1 Elements suitable for direct adaptation
Five elements have high transferability. First, Hungarian universities can establish a permanent AI education and governance centre or network. Its mandate should combine pedagogy, legal compliance, information security, procurement and staff development. Second, they can introduce a common scale for assignment-level AI permissions. Third, they can adopt proportional disclosure templates. Fourth, they can build staged AI-literacy programmes for different roles. Fifth, they can provide approved institutional services instead of leaving every user to select external tools.
These measures do not require a change in academic values. They make existing values operational. Academic freedom becomes more usable when teachers receive common resources. Data protection becomes more credible when secure tools are available. Academic integrity becomes easier to enforce when students know what to disclose.
5.2 Elements requiring legal and organizational adaptation
The EU environment requires explicit attention to lawful processing, purpose limitation, data minimization, transparency and rights. An institutional platform must document controller and processor roles, retention periods, security measures and international data transfers. Procurement must assess both the model provider and the way the university deploys the system.
The AI Act requires use-case classification. A chatbot that explains course material differs from a system that evaluates learning outcomes or influences admission. The latter may trigger high-risk obligations, including risk management, documentation, human oversight, accuracy and monitoring. Hungarian adaptation must therefore connect educational design with legal classification at the level of the concrete system and purpose.22
Organizational culture also matters. Hungarian universities often protect faculty autonomy and operate through distributed decision-making. A central AI unit should supply infrastructure and common safeguards while allowing faculties to define disciplinary rules. A network model may fit this environment better than a strongly centralized centre: a small central team can coordinate faculty representatives and maintain shared services.
5.3 Mutual learning
The comparison also identifies lessons for Fudan. European practice foregrounds data-subject rights, documented legal bases, proportionality and procedural accountability. These concepts can strengthen trust in university AI systems, especially when Fudan collaborates internationally or processes data across jurisdictions. Hungarian institutions also demonstrate how disciplinary communities can shape rules close to teaching practice.
Fudan contributes the stronger model of institutional integration. Hungarian universities contribute a mature rights-based vocabulary and experience with distributed academic governance. A combined model would connect strategic capacity with legal safeguards, and shared infrastructure with disciplinary autonomy.
6. An Adaptation Framework for Hungarian Universities
A Hungarian university can translate the comparative findings into a six-part programme.
1. Establish responsibility. The rectorate should appoint an institution-wide AI governance body with academic, legal, technical, library, quality-assurance and student representation. The body should publish a mandate, decision process and annual work programme.
2. Create a common usage framework. The university should define baseline prohibitions and a four-level permission scale for courses and assignments. Teachers should state the applicable level in the syllabus and task description.
3. Redesign assessment. Faculties should identify high-impact courses and develop staged submissions, oral components, reflective statements and authentic tasks. The institution should prohibit the use of AI-detection scores as sole evidence of misconduct.
4. Introduce transparent declarations. The university should provide short and extended templates. The short form should name the tool, purpose and student's contribution. The extended form should document prompts, verification and revision where the task or risk justifies it.
5. Build competence. All staff should receive baseline AI-literacy training. Teachers, researchers, administrators and technical staff should receive role-specific modules. Advanced users should form communities of practice and develop reusable course designs.
6. Provide secure infrastructure. The university should offer approved AI services, publish a data-classification guide, review providers, document data flows and integrate authentication and support. It should test services through limited pilots before institution-wide deployment.
Implementation should proceed through iterative governance. The institution can begin with a small number of faculties, collect declarations and user feedback, examine incidents and revise rules. Annual reporting should cover training, approved tools, assessment redesign, security events and user experience. This learning cycle reflects one of Fudan's most important insights: AI governance must operate as a living institutional system.
7. Conclusion
Fudan University and Hungarian universities share the same fundamental goals: they seek to preserve academic integrity, strengthen learning, protect information and prepare students for AI-enabled professional life. They differ in how they connect these goals with institutional capacity.
Fudan has built the more integrated model. Its guidance forms part of a system that includes a dedicated centre, staged competence development, procedural disclosure, assessment reform and technological infrastructure. The university treats AI governance as an organizational and pedagogical transformation.
Hungarian universities have developed substantial normative expertise. Their rules increasingly permit responsible assistance, require transparency and protect independent intellectual work. The EU legal environment adds strong safeguards for personal data, fundamental rights, human oversight and accountable deployment. Hungarian practice remains fragmented across institutions and faculties, and teachers often carry implementation responsibility without a comparable platform or permanent support structure.
The principal transferable lesson is institutional integration. Hungarian universities can join rules, training, assessment, data governance and infrastructure through a coordinated programme. They can do this while retaining faculty autonomy and applying European legal safeguards to each concrete use case. The resulting model would not reproduce Fudan. It would translate Fudan's systemic logic into a Hungarian and European institutional form.
The future of university AI governance will depend on this capacity for translation. General ethical principles acquire force when institutions assign responsibility, provide secure tools, redesign assessment and develop human competence. Universities that build these connections can use AI to expand intellectual agency while preserving the standards that give academic work its value.