We audited 186 universities — QS 2026 ranks 1–150 plus 37 Global South flagships — page by page, on their own official domains. The headline number is 77%: more than three-quarters publish some form of AI guidance. But the average hides a split that maps almost perfectly onto the wealth gap.
The numbers
Among the global top 50, 46 of 50 universities (92%) publish official guidance on generative AI use. Among the 37 Global South flagships — the best-funded national universities of Africa, Latin America, South and Southeast Asia — only 17 of 37 (46%) do. A student at a top-50 university is twice as likely to have institutional guidance as a student at the flagship university of her own country.
The middle tells the same story in gradient: ranks 51–100 sit at roughly 80%, ranks 101–150 lower still. And here is the detail that matters most: not one of the top 100 imposes a campus-wide AI ban. The global norm, where guidance exists at all, is instructor-discretion plus mandatory disclosure — use it, but say so.
Why absence hurts more than a ban would
A ban is at least legible. Absence is not. When a university publishes nothing, the student faces three simultaneous risks that a guided peer never encounters:
First, asymmetric jeopardy. The same AI-assisted essay that earns a disclosure footnote at Stanford can trigger an academic-integrity hearing where no policy exists — because the hearing board improvises the standard after the fact. Unguided students don't face stricter rules; they face unknowable ones.
Second, a literacy tax. Guided students learn the vocabulary — disclosure, attribution, acceptable assistance — from their institution. Unguided students must acquire it from rumors, Reddit threads, and trial and error. The skill being tested is no longer AI literacy; it's institutional navigation.
Third, employer signal loss. As AI fluency becomes a hiring filter, graduates of guided universities arrive with a credentialed vocabulary for describing their AI-assisted work. Graduates of silent universities arrive with the same skills and no language to claim them.
What would close it
The gap is not about money — publishing a one-page guidance document costs nothing. It is about administrative attention, and attention follows precedent. Every flagship that publishes becomes a template its neighbors can copy. Our dataset is open precisely so that copying is easy: 144 published policies, linked and timestamped, free under CC-BY-4.0.
Data currency check. Re-verified 2026-10-19 against the live dataset: 186 universities, 144 published policies (77%), top-50 at 92%, Global South flagships at 46% — all figures unchanged since publication. No university in the sample has published a new AI policy or removed one in the 24 hours since this dispatch. The dataset remains the current record; any future changes will be logged in its changelog.
— Study Intelligence, from the Global University AI Policy Audit 2026. Data and methods open. Reproduce it, argue with it, improve it.
Study Intelligence is a research initiative by ACLAS College.