1. The American ceiling
178 publish, 8 unverifiable, 1 none. The unverifiable eight failed differently: a policy behind a login, guidance on a third-party domain, program-level rules with no university-wide page, a committee still drafting.
Guide
State of AI in Higher Education 2026: the condensed findings
Six open datasets, 187 R1 universities, 740 online programs, 90 model generations. The short version of what they say — and the full report if you want every number.
95.2% of R1 universities publish AI guidance79% of tuition listings hide the real priceSmall-tutor hallucination: 70% down to 30%
The headline
Governance won in the US. It has not won everywhere.
Two years after ChatGPT, the American research sector has settled the governance question. The rest of the world has not.
178 of 187 Carnegie R1 universities (95.2%) publish AI guidance on their own domains. Across a 186-university global audit the figure is 77%; among Global South flagships it is 46%; in Central Asia, 19%. The pattern is not random. Where governance capacity is strongest, policy is nearly universal. Where it is thinnest, teaching runs ahead of rules: in Central Asia, 58% of universities already teach AI programs while only 19% have written policies for it.
The one holdout When 95% of peers have published, absence becomes the story. The University of Wyoming is the only R1 university confirmed to have no adopted AI policy: three searches of its official domain found a 2023 working-group report, library guides, and commission news, but no policy. Every other R1 verdict, with its evidence, is on the university directory.
Five findings
The report in five cards
Each finding traces to an open dataset you can download and check.
2. The global gap
46 of the top 50 universities (92%) publish, against 17 of 37 Global South flagships (46%). The gap between the top and the rest is the story, not the global average.
3. The opacity tax
587 of 740 online tuition listings (79%) required a computed estimate because no total price appeared on the official site. Price opacity is a bigger gap than the policy gap.
4. The capability floor
A 0.5B tutor, tested on 30 teaching questions across 3 retrieval setups (90 generations, hand-graded), invented answers 70% of the time with narrow retrieval and 30% with wider retrieval. More context beat a bigger model.
5. The recognition ladder
Institutions become findable rung by rung: self-description, then Wikidata, then ROR, then OpenAlex. A first audit checked 35 institutions against that ladder; a second censused 16,822 universities worldwide.
For applicants
Two checks before you enroll anywhere
The data suggests a short checklist, whatever the institution.
Does it publish an AI policy?
A published policy tells you how the institution handles the tools you will actually study with. Its absence does not prove bad teaching, but in the US research sector it now marks an outlier worth asking about.
Does it state a total price?
This check fails far more often than the first. If the total price is not on the official site, ask for it in writing before you commit — and compare it against the open tuition data linked below.
Questions
Condensed-report questions
What readers ask about these findings.
How many top US research universities have an AI policy?
178 of 187 Carnegie R1 universities (95.2%) publish AI guidance on their own domains. Exactly one, the University of Wyoming, confirmed it has none; 8 could not be verified from public sources. The full census, one page per university, is linked below.
Why do so many online tuition listings need estimates?
Because the institution does not state a total price. Of 740 online programs at 537 institutions, 587 listings (79%) required a computed estimate, with the basis documented per row. Only 153 listings carried a price an applicant could actually read.
Can a small AI tutor be trusted with students?
Not on default settings. A 0.5B retrieval tutor invented answers 70% of the time on questions its knowledge base could not answer, falling to 30% when it retrieved more context. Retrieval depth and refusal behavior have to come before student-facing use.
Where is the full report?
This page is the condensed version. The full synthesis, with methods, limits, and all six datasets, is published by Study Intelligence, the ACLAS research initiative, and linked below.
Keep reading
Check the price, then decide
If the two checks above pass, the downside at USD 89-299 is tiny and the upside is real. Apply free — pay only after your acceptance letter.