AI and ethics: Investigating the first policy responses of higher education institutions to the challenge of generative AI

The rapid emergence of generative AI, especially with the release of ChatGPT in late 2022, triggered wide public attention and concern, including calls for pauses in advanced AI development. As competition in AI accelerated, higher education institutions (HEIs) faced urgent questions about how to adapt teaching, learning, and assessment. This study aims to reduce knowledge gaps by providing a global snapshot of how universities responded within six to eight months of ChatGPT’s arrival. The central research question is: what expectations and guidelines did early university policies introduce to ensure informed, transparent, responsible, and ethical use of generative AI by students and teachers? Focusing on leading institutions likely to shape best practices, the study analyzes public-facing policy and guidance documents from universities ranked in the Shanghai Ranking top 500 that had published early AI-related responses. The investigation is organized around ethical principles derived from prominent international sources (UN, EU, OECD).

The paper surveys emerging scholarship on generative AI in higher education (HE). Historically, educational technology has cycled through waves of optimism and concern; LLM-based chatbots are the latest disruptive force. Potential benefits cited include personalized feedback and tutoring, adaptive learning, and automation of instructional tasks. However, significant challenges are highlighted: risks to academic integrity and assessment validity; unreliable AI-detection tools; ethical concerns around privacy, bias, fairness, accountability, transparency; over-reliance and potential erosion of learner agency; unequal access and digital divides. Frameworks such as FATE (Fairness, Accountability, Transparency, Ethics) and policy proposals (e.g., Chan’s pedagogical/governance/operational dimensions) inform discussions, but prior literature often focuses narrowly on ChatGPT and assessment, with limited attention to actual institutional practices at scale. This motivates a systematic review of universities’ early policy responses aligned to widely recognized ethical principles.

Research objective: to review how leading HEIs reacted to generative AI’s arrival and whether key ethical principles were reflected in early policies and guidance. Design: qualitative, directed content analysis in three steps. Step 1: Identify ethical ideals from internationally respected AI governance documents and translate them into the HE context. Five sources were selected: EU Artificial Intelligence Act proposal (European Commission, 2021), European Parliament Committee on Culture and Education report (2021), EU High-Level Expert Group on AI Ethics Guidelines for Trustworthy AI (2019), UNESCO Recommendation on the Ethics of AI (2022), and OECD Recommendation on AI (2019). From these, four ethical clusters relevant to HE practice were synthesized: (a) Accountability and responsibility, (b) Human agency and oversight, (c) Transparency and explainability, (d) Inclusiveness and diversity. Step 2: Case selection and data collection. From May–July 2023, the authors (as part of a university AI committee) screened over 150 institutional cases globally and narrowed to universities in the Shanghai Ranking top 500 that had publicly available, AI-relevant policies or guides with identifiable ethical considerations. The final sample included 30 universities across six continents. Sources included Codes of Ethics, Academic Regulations, Codes of Practice and Procedure, and guidelines for students and teachers. Step 3: Analysis. Both authors independently annotated documents for references to the four ethical clusters, then compared and synthesized findings to identify convergences, divergences, and emerging practices. The study focuses on AI use in education (not AI development) and emphasizes descriptive mapping of first responses rather than prescriptive claims.

- Sample and scope: 30 universities from the Shanghai Ranking top 500 with early, publicly available generative AI guidance (May–July 2023), spanning six continents. - Ethical clusters distilled from international documents: (1) Accountability and responsibility; (2) Human agency and oversight; (3) Transparency and explainability; (4) Inclusiveness and diversity. - Accountability and responsibility: Core imperative across universities is that students’ assessed work must reflect their own learning; AI cannot bear legal or moral responsibility. Unauthorized submission of AI-generated work is treated as plagiarism, cheating, or misuse unless explicitly permitted. Institutions re-affirm existing academic integrity frameworks; some adopt flexible stances (e.g., Tokyo Institute of Technology emphasizes ethical independence; York University allows use only if an instructor explicitly permits; Oxford flags unauthorized AI use as a serious offense). - Human agency and oversight: Given unreliable AI-detection tools and privacy concerns, many universities discourage their use (e.g., National Taiwan University, Toronto, Waterloo, Miami, UNAM, Yale). ETH Zürich warns of fairness issues; Cape Town notes disproportionate false flags for non-native writers. Where used, detection tools are limited to formative insight and not punitive proof (e.g., Boston University). Preventive redesign of assessments (oral/practical/in-person exams, personalization, in-class drafts, authenticity interviews) emerges as preferred oversight, with soft, dialogue-based responses to suspected misuse. - Transparency and explainability: A bottom-up model prevails, emphasizing instructors’ autonomy to set course-level AI rules, coupled with clear syllabus statements. Many institutions provide tiered frameworks: 3-level (e.g., Colorado State; Macquarie; Monash) or 4-level (e.g., Chinese University of Hong Kong; University of Delaware) policies ranging from prohibition to full use with/without acknowledgement. Instructors are urged to state permitted scope and attribution requirements in syllabi. - Inclusiveness and diversity: Institutions promote equitable access and participation by offering centrally supported examples and resources (e.g., Waterloo, Pittsburgh, Monash), scaffolding approaches (Columbia), accessibility requirements if AI use is mandated (Humboldt-Universität zu Berlin), recognition of connectivity/device divides (Cape Town), and broader inclusion initiatives (e.g., TU Berlin’s “Inclusive Digitalisation” module; Cambridge’s AI-ideas projects). Community-building and AI literacy efforts (e.g., UNAM) aim to reduce disparities. - Notable data point: Among Germany’s 100 largest universities (Solis, 2023) early stances were heterogeneous: 2% general prohibition, 23% partial permission, 12% general permission, 63% none or vague guidance—illustrating fluid, evolving policy landscapes. Overall: Top-down ethical imperatives (human accountability) coexist with bottom-up instructional autonomy, with best practices favoring preventive assessment design, soft sanctions, transparent syllabus communication, and inclusive central supports.

The findings directly address the research question by mapping how early university policies operationalized widely accepted AI ethics principles in educational contexts. Institutions largely reframed existing academic integrity rules to encompass generative AI while empowering instructors to tailor classroom practices. This balance advances responsible adoption by ensuring human accountability and preserving learner agency, while transparency via syllabus policies clarifies expectations and supports informed student choices. The emphasis on preventive assessment redesign, rather than punitive detection, acknowledges current technical limits of AI detection and aligns with fairness and privacy concerns. Inclusivity-focused measures (e.g., access to tools, scaffolding, AI literacy resources) help mitigate digital divides and promote equitable learning opportunities. Collectively, these responses represent pragmatic, ethically grounded pathways for integrating generative AI in HE, providing a foundation for continuous policy refinement as the technology evolves.

The study synthesizes international AI ethics guidance into four HE-relevant ethical clusters and documents how 30 leading universities’ first responses reflected these principles. Key contributions: (1) Identification of a central ethical imperative—student work must represent individual learning and humans remain morally and legally accountable; (2) Documentation of a dual strategy—top-down integrity rules combined with bottom-up instructor autonomy and clear communication; (3) Highlighting effective oversight practices emphasizing assessment redesign and dialogue-based approaches over unreliable AI detection; (4) Emphasis on inclusivity through centrally provided resources, access arrangements, and AI literacy. Future research should (a) extend sampling beyond early adopters and revisit institutions longitudinally to capture policy evolution, (b) examine research-related uses of AI within universities, and (c) evaluate the educational impact of specific policy instruments (e.g., tiered syllabus frameworks, authenticity interviews, inclusive access provisions).

The scope is limited to pedagogical/teaching contexts and excludes university research uses of AI. The sample captures early adopters with publicly available documents among Shanghai top 500 institutions and may not generalize to all HEIs. Findings reflect a specific time window (May–July 2023) during a period of rapid technological and policy change. The analysis is descriptive of first responses rather than evaluative of long-term effectiveness.