The scale of this survey makes the gap hard to dismiss
The Digital Education Council, a global higher education research and policy organization, published its AI in Higher Education Global Survey 2026 this month, drawing on 45,398 responses from 35 countries: 27,284 students and 18,114 faculty members. That is a large enough sample to move past anecdote and into something institutions can actually plan around, and the headline finding is not that AI adoption in higher education is stalling. It is that adoption is outpacing the institutional infrastructure meant to guide it, and the gap is widest in exactly the markets, the US and Canada, that tend to set the pace for enterprise AI adoption more broadly.
Some of the individual numbers are stark on their own. Only 29 percent of students globally believe their instructors are well equipped to guide AI use in coursework, and that figure drops to just 17 percent among US and Canadian students specifically, the lowest confidence level in the entire dataset. Meanwhile 15 percent of students report AI integration across many of their courses, and another 43 percent see it in a few courses, meaning a majority of students are already encountering AI in the classroom regardless of whether their institution has a coherent policy in place.
Faculty confidence is falling in exactly the wrong market
The most counterintuitive result in the survey is regional: faculty intent to use AI in teaching in the US and Canada fell 9 percentage points from 2025 to 67 percent in 2026, even as majorities of faculty across APAC, EMEA, and Latin America maintained or grew their adoption intent over the same period. That is a reversal worth sitting with. The region most associated with AI product development and enterprise AI spending is the region where the faculty closest to actually teaching with these tools are pulling back, not leaning in.
The survey's own data suggests why. 64 percent of faculty globally report participating in some form of AI literacy training, which sounds like meaningful progress until it is compared against the 15 percent classroom integration figure and the 17 percent student confidence figure in North America specifically. Training completion and actual comfort deploying that training in a live classroom are clearly two different things, and the gap between them appears to be where US and Canadian faculty confidence is going to erode rather than build.
Students are more worried about each other than about the technology
One of the more striking findings concerns peer behavior rather than the AI tools themselves. 60 percent of students globally say they worry about peers misusing AI, and that figure climbs to 73 percent among US and Canadian students, the highest regional figure in the survey. That is a trust and integrity concern as much as a technology concern, and it is showing up in policy preference: 43 percent of US and Canadian students now say they support an institution-wide AI ban, a position that sits in tension with the reality that most of those same students are already using AI tools regularly in coursework.
There is a workplace-relevance signal buried in the same dataset that deserves equal attention. 72 percent of students globally say their assessments do not feel aligned with the AI-enabled skills an actual workplace will expect from them, and 37 percent doubt their program of study is preparing them for an AI-driven future at all. Institutions can read that as a curriculum problem, but employers evaluating new graduate pipelines should read it as a direct signal: the students entering the workforce over the next several years do not, in large numbers, believe their own coursework prepared them to use AI the way a job will actually require.
Why this matters beyond the campus
Enterprise technology leaders have two direct reasons to pay attention to this survey, and neither requires a stake in higher education policy. The first is the talent pipeline. If nearly three-quarters of US and Canadian students distrust their peers' AI use and more than a third doubt their program's relevance to an AI-driven career, employers hiring recent graduates should expect a wide and uneven range of actual AI fluency arriving in entry-level roles, regardless of what a resume or a degree program implies. Onboarding and early-career training programs built assuming baseline AI competence from new hires are building on an assumption this survey does not support.
The second reason is structural, and it is the one enterprise leaders should find most familiar. The same pattern this survey documents in higher education, adoption racing ahead of governance, training completion not translating into confident practice, and a widening trust gap around how peers are actually using the tools, is the identical shape of the AI governance problem showing up inside large enterprises right now. Universities are simply running the experiment at a scale and with a research apparatus, 45,000 respondents across 35 countries, that most corporate AI governance functions do not have access to about their own workforce.
The governance move this data actually supports
The obvious but wrong response to this survey is to conclude that more AI literacy training will close the gap, since 64 percent of faculty have already completed some form of it without meaningfully moving the classroom integration or student confidence numbers. The more useful response is to treat training completion and governance readiness as two separate metrics that need to be tracked and funded separately, because this data shows clearly that improving one does not automatically improve the other.
For any organization, campus or enterprise, currently measuring AI governance maturity purely by training completion rates or policy documents published, this survey is evidence that those are lagging, not leading, indicators. The leading indicators are closer to the ones the Digital Education Council actually measured: whether the people doing the work believe the guidance they received translates to their specific daily use case, and whether they trust their peers to use the tools responsibly under the same rules. Those are harder numbers to collect than a training completion rate, and this survey is a reasonable argument that they are the numbers that actually matter.
