AI Case Studies  /  Case study

How Universities Are Actually Using AI in 2026: What EDUCAUSE Data and Named Campuses Show

Higher education has moved past the pilot stage into system-wide rollouts, in-house tools built behind privacy walls, and graduation requirements. The verified record from EDUCAUSE research and named institutions shows where the real adoption sits, and why the rules have not kept up.

By Anthony Guerriero  ·  Reviewed by The Leveraged Years Editorial Desk  ·  Published June 28, 2026

By 2026, AI in higher education is no longer experimental. EDUCAUSE research finds that 92 percent of surveyed institutions have a work-related AI strategy and 94 percent of higher-ed employees have used AI tools for work in the past six months, yet only 54 percent know what policies govern that use. The deployments are concrete: Cal State put ChatGPT Edu in front of 460,000 students across 23 campuses, the University of Michigan built its own privacy-walled generative AI suite, and Ohio State now requires every undergraduate to graduate AI-fluent in their field. The defining tension of the year is the gap between how much faculty and staff use these tools and how little they know about the rules.

From pilots to strategy: what the EDUCAUSE numbers actually say

Picture a campus where almost everyone uses AI at work and almost half cannot say what they are allowed to do with it. That is not a hypothetical. In January 2026, EDUCAUSE published "The Impact of AI on Work in Higher Education," a study of 1,960 higher-ed employees fielded with the Association for Institutional Research, NACUBO, and CUPA-HR. The top-line number is hard to argue with. Ninety-two percent of respondents said their institution now has a work-related AI strategy, and only 5 percent reported none. Just 5 percent of institutions are discouraging or prohibiting AI use. The posture has flipped from caution to adoption. Eighty-one percent of employees report enthusiasm or a mix of caution and enthusiasm, and only 17 percent are purely cautious.

Use has caught up to strategy. Ninety-four percent of respondents had used AI tools for work in the prior six months, and among recent users, 73 percent reach for them daily or weekly. The everyday applications are unglamorous and revealing: brainstorming (63 percent), drafting emails (62 percent), summarizing documents and meetings (61 percent), and proofreading (56 percent). This is not a story about robot tutors replacing professors. It is a story about the academic workforce quietly folding AI into the daily grind.

The same study exposes the problem that should keep provosts and CIOs awake. While 94 percent are using these tools, only 54 percent know of any policies or guidelines governing that use, which means nearly half are working blind to the rules. Fifty-six percent have used AI tools their institution never provided, the textbook definition of shadow AI and the source of most data-exposure risk. Only 13 percent of institutions are measuring return on investment, and only 11 percent of employees are actually required to use AI, against 86 percent who want to keep using it. Adoption is racing ahead of the rules, and that gap is where the institutional risk now lives.

The earlier 2025 EDUCAUSE AI Landscape Study, "Into the Digital AI Divide," set up this moment. As reported by GovTech, that survey found 57 percent of institutions treating AI as a strategic priority, up from 49 percent a year earlier, with 54 percent already using AI for curriculum design and 52 percent for administrative automation. Acceptable-use policies had climbed to 39 percent of institutions from 23 percent the year before, real progress that still leaves a majority without one. Only 9 percent of respondents said their cybersecurity and privacy policies adequately addressed AI risk. EDUCAUSE named the central tension a "digital AI divide," with larger and better-resourced institutions far more likely to have policies, IT support, and budget. That divide did not close in 2026. It hardened.

The system-scale deployments: ASU and Cal State

To see where strategy turns concrete, start with the OpenAI partnerships. Arizona State University was first. ASU announced its collaboration with OpenAI on January 18, 2024, becoming the first university to deploy ChatGPT Enterprise, a move CNBC flagged as OpenAI's first university partnership. ASU did not just buy licenses. Its Enterprise Technology unit stood up an "AI Acceleration" team to build custom tools, then ran an internal "AI Innovation Challenge" that drew more than 175 proposals in its first round, of which 105 were accepted. By September 2025, ASU had expanded the program to offer ChatGPT Edu with GPT-5 to every student, faculty, and staff member at no cost, and reported that the platform had activated more than 500 projects, supplemented by an in-house toolkit called CreateAI. The ASU model rewards study precisely because it pairs a commercial platform with the internal muscle to build on top of it.

California State University took the same idea and scaled it past anything seen before. On February 4, 2025, the CSU system announced it was deploying ChatGPT Edu to more than 460,000 students and 63,000 faculty and staff across all 23 campuses. CSU described itself as the first AI-powered university system in the United States, and OpenAI called it the largest deployment of ChatGPT to date. Chancellor Mildred García framed it as an equity move, putting the same tools in the hands of a student body that is heavily first-generation and lower-income. The scale is the point. A single decision put enterprise-grade AI in front of nearly half a million students at once, which is a different category of institutional commitment than a departmental pilot.

The build-it-yourself camp: Michigan, Vanderbilt, and the walled garden

Not every leading institution outsourced the platform. The University of Michigan went the other way. In August 2023, U-M launched a custom suite of generative AI tools, U-M GPT, Maizey, and a developer-facing GPT Toolkit, and claims to be the first university in the world to provide such a suite. The design choice that matters is the architecture. Michigan built the tools with Microsoft inside what it calls a walled-off Michigan environment, where prompts and data are not used to train outside models. By early 2024 the platform was serving roughly 15,000 users a day. For institutions worried about data exposure, Michigan is the reference implementation: keep the data inside the fence, offer the free general tool, U-M GPT, to everyone, and charge departments for the heavier custom builds.

Vanderbilt followed a similar instinct, and it corrects the assumption that everyone simply bought ChatGPT. Vanderbilt's flagship is Amplify, a custom secure generative-AI sandbox it built and opened to all faculty, staff, and students in November 2024. Only in January 2026 did Vanderbilt add ChatGPT Edu alongside Amplify 2.0, adopting an explicitly multi-tool strategy and reporting that more than 70 percent of its campus community uses these tools regularly. The lesson for IT leaders is that the in-house option and the commercial option are not mutually exclusive, and the most committed campuses run both. Harvard and MIT sit in the same camp on the infrastructure question. Harvard operates a walled-off HUIT AI Sandbox that pools multiple models, and MIT rolled out Parley, its own secure multi-model platform spanning Llama, Gemini, Claude, and OpenAI, in March 2026.

Curriculum and the research stack: Florida and Ohio State

The deployments above are about access. Two institutions show what happens when AI gets baked into the academic mission itself. The University of Florida made the earliest big bet, announcing a $70 million public-private partnership in July 2020 anchored by a $25 million gift from UF alumnus and NVIDIA co-founder Chris Malachowsky, $25 million in kind from NVIDIA, and $20 million from the university. The result, HiPerGator AI, runs on an NVIDIA DGX SuperPOD with 1,120 A100 GPUs and was described as the fastest supercomputer owned and operated by a university in the United States. UF then wrapped a curriculum around the hardware, building toward what it calls an "AI University": more than 200 AI courses, a university-wide AI Certificate, more than 100 new AI faculty hires across all 16 colleges, and roughly 12,000 students a year taking the coursework.

Ohio State pushed the curricular idea to its logical end. On June 4, 2025, the university announced its "AI Fluency" initiative, declaring that every undergraduate beginning with the Class of 2029 will graduate AI-fluent in their field. The requirement is embedded in general education, with a required Launch Seminar and a new course on generative AI, framed by President Ted Carter as part of an "Education for Citizenship 2035" agenda. It is not a registrar checkbox so much as a foundational expectation woven through the degree. Whether the execution matches the announcement is a fair question, but the signal is unmistakable. A flagship public university now treats AI literacy as a graduation-level competency on par with writing.

The oldest lesson: AI for advising and tutoring

The flashiest 2026 deployments rest on a foundation that predates ChatGPT. Georgia State University has been running an AI advising chatbot, Pounce, since 2016. The documented results are why every dean now takes administrative AI seriously. A randomized controlled trial tied to the chatbot found a 3.3 percent enrollment increase and a 21.4 percent reduction in summer melt, the phenomenon where admitted students fail to show up in the fall. GSU officials have separately reported cutting summer melt from 19 percent to 9 percent, with Pounce interacting with first-year students roughly 185,000 times in a single summer while fewer than 1 percent of messages required staff escalation. The teaching side has its own long-running example in Georgia Tech's Jill Watson, the AI teaching assistant that debuted in 2016 and now, integrated with current models, reports 78.7 percent answer accuracy in real classrooms.

What ties the whole 2026 picture together is the gap between students and the institutions serving them. Independent survey data from the Digital Education Council's 2024 global study of 3,839 students found that 86 percent already use AI in their studies, with two-thirds using ChatGPT specifically. Faculty trail badly. Tyton Partners' "Time for Class" work, reported by Inside Higher Ed, put regular instructor use at roughly 36 percent against three in five students. The institutions profiled here, from Cal State's half-million-seat rollout to Ohio State's fluency mandate, are best understood as attempts to close that gap on the institution's terms, with governance and data protection attached, before students close it on their own with whatever tool they can reach.

Deployment details are institution-specific and current as of mid-2026; this is informational analysis, not implementation or legal guidance.

Browse all AI Case Studies

Deployment details are institution-specific and current as of mid-2026; this is informational analysis, not implementation or legal guidance.