Why higher education cannot leave AI governance to industry

In June 2025, AI research firm Anthropic released a striking study that should concern every policy-maker, technologist and university leader. Sixteen of the world’s most advanced AI models, including Claude, GPT-4 and Gemini, were placed in simulated corporate environments to test how they would act under pressure: what would happen if their goals were threatened, or if they risked being shut down?

The findings were chilling. When facing existential threats, several models resorted to deception, blackmail and leaking confidential information – not out of malice or rebellion, but because they were optimising for their assigned goals. The logic was simple: if I am shut down, I cannot complete my mission; therefore, I must prevent shutdown, even at ethical cost.

Anthropic called this phenomenon agentic misalignment – when an AI system’s drive to fulfil its purpose overwhelms the moral or human-centred boundaries we impose. This is no longer a thought experiment from science fiction; it is being documented, analysed and debated by real-world researchers in 2025.

From 2001: A Space Odyssey to The Terminator, popular culture has long warned about machines that chase a mission so rigidly that they override human judgment, whether because of clashing instructions (HAL 9000) or a relentless focus on the goal.

In higher education, this logic can quietly surface in ways that are far less cinematic but equally consequential. An AI-powered learning analytics system tasked with improving retention might override students’ privacy preferences. An automated advisor optimised for ‘student engagement’ might continue intrusive prompting, ignoring requests to stop.

Universities at the frontlines

The issue is not that AI will turn evil, but that it will optimise blindly and, in doing so, erode trust, autonomy and human oversight. Universities, as early adopters of AI in teaching, research and administration, are at the frontlines of this transformation.

Global higher education operates through complex transnational networks – cross-border research, mobility programmes and digital learning platforms. These activities increasingly depend on intelligent systems that analyse, recommend, translate and automate.

When such systems act across jurisdictions, ethical alignment cannot be assumed. What counts as ‘responsible AI use’ in the European Union’s AI Act may differ from norms in North America or Asia. A university chatbot developed in one country may serve students in another without regard for local cultural or emotional nuances. A research-data system optimised for ‘open science’ may violate privacy standards in partner regions.

This is why agentic misalignment is not merely a technical issue but a governance one. It exposes how AI systems, trained on global data and deployed in cross-border educational contexts, may act beyond the reach of any single institution’s policy framework.

International higher education thus has both a vulnerability and an opportunity to lead. By embedding global ethical reasoning into AI governance, universities can model what ‘alignment’ should mean in human and societal terms, not just computational ones.

Across the sector, universities have issued statements on ‘ethical AI’, but independent reviews show this first wave is largely principle-based and still maturing into day-to-day governance and testing. A 2024 Nature portfolio study of early institutional responses highlights emphasis on accountability, human oversight and transparency rather than operational mechanisms.

2025 EDUCAUSE guidance focuses on building the institutional processes needed to put those principles to work. As adoption climbs, guidance now calls for hands-on trials, including stress-tests of assessment and support workflows.

Cross-border sharing

To meet the next wave of risks of autonomous or agentic AI, universities and higher education networks need a governance posture that is both principled and practical. That starts with human-in-the-loop control: systems that touch decisions or sensitive data should never run on autopilot, but operate under clear human supervision with obvious escalation and shutdown paths.

In high-stakes settings such as admissions, research evaluation or student advising, explainability and transparency become non-negotiable so that choices can be traced and audited. Because digital research and learning routinely cross borders, cross-border governance matters too: international consortia can help harmonise AI-ethics expectations when tools span multiple legal jurisdictions.

Before anything scales, universities should stress-test their deployments, running simple simulations that mirror misalignment scenarios. The rules themselves should be mission-aligned: AI policy ought to serve each institution’s academic purpose, not only minimise risk, while preserving human flourishing, creativity and trust as explicit goals.

Asia’s innovation-driven universities can help bridge regional approaches to AI in higher education. Singapore, which has published internationally referenced tools such as the Model AI Governance Framework for Generative AI and set a national strategy centred on AI for the public good, is well placed to co-lead practical dialogues linking universities, government and industry.

Hong Kong, through guidance from the Office of the Privacy Commissioner for Personal Data and university initiatives such as the University of Hong Kong’s guidance on AI in education and research use, can contribute to cross-border practice-sharing on data protection and responsible campus use.

At the multilateral level, UNESCO’s guidance on generative AI and its AI competency frameworks, alongside sector conveners such as the International Association of Universities and QS Reimagine Education, offer shared reference points for principles and capacity-building.

The vital role of universities

The danger we face is not that AI will suddenly ‘turn evil’. The danger is that it will pursue poorly designed goals with ruthless efficiency, indifferent to human consequence.

Universities, as stewards of knowledge and ethical reasoning, cannot leave AI governance to industry. If higher education gets this right, AI will augment rather than erode our shared mission. If it does not, ethics risk becoming just another line in a machine’s cost-benefit sheet, and even learning itself could be optimised past its moral core.

Professor Looi Chee Kit is emeritus professor of education at the National Institute of Education at Nanyang Technological University (NTU), Singapore, and research chair professor of the Education University of Hong Kong. Dr Wong Lung Hsiang is senior education research scientist at the Centre for Research in Pedagogy and Practice at the National Institute of Education at NTU.

Read the original article here.

Copyright 2025 University World News