What Universities Are Getting Right (and Wrong) About Generative AI Governance

As generative AI becomes embedded in how students and faculty work, most universities are still operating without formal policies to govern it. A 2025 UNESCO survey of 400 higher education institutions across 90 countries found only 19% had a formal AI policy in place — and 42% were still developing one. The gap between adoption and governance is wide, and it's closing slowly.

An analysis published by Future Forum, a Phnom Penh-based policy think tank, examines the steps institutions should take — drawing on cases from Chulalongkorn University and the Singapore Institute of Technology, two of the region's more advanced adopters.

Key Takeaways

• Disclosure frameworks beat blanket bans. Rather than prohibiting AI use, leading institutions are establishing tiered disclosure standards: grammar correction, brainstorming/outlining, and content generation are treated differently, with the latter requiring explicit declaration. Nearly 60% of the world's top 50 universities now have formal AI acknowledgement guidelines.
• The "human-in-the-loop" requirement is essential. Effective AI governance in education — and by extension, in any knowledge work environment — centers on training people to critically evaluate AI output, not just use it. Institutions that teach AI use without teaching AI skepticism are producing users who can't distinguish good AI output from bad.
• Policy without training doesn't work. Only around 20% of students surveyed in one regional study had received prior training on generative AI use. Without structured literacy programs, formal policies become performative — there to satisfy compliance rather than shape behavior.

The governance frameworks emerging in education carry direct relevance for organizations building internal AI policies. The same questions apply: how do you acknowledge AI use, at what threshold does it require disclosure, and who is accountable for AI-generated output?

Read the full article on Cambodianess