Governance Without Readiness: How 53 American Universities Responded to Generative AI
Abstract American universities responded to generative AI primarily by governing it.
Abstract American universities responded to generative AI primarily by governing it. This paper examines how 53 institutions built that response and what they did not build alongside it. Using public-source institutional profiling across a consistent ten-section framework, the study documents AI governance and workforce readiness activity at 23 nationally ranked peers of William & Mary and 30 Virginia four-year institutions, with an evidence snapshot date of March 2026.
The findings reveal a structural asymmetry. Ninety-eight percent of institutions formed AI governance committees. Ninety-two percent revised academic integrity codes. Eighty-five percent provisioned enterprise AI tools. But only 7.5 percent documented employer outreach regarding AI expectations for graduates, and 5.7 percent included career services in their AI governance structures. Of 53 institutions, one originated its AI strategy from employer input. Twenty-nine were governance-dominant with no visible readiness infrastructure.
Drawing on Weick’s loose coupling, DiMaggio and Powell’s institutional isomorphism, and Cohen, March, and Olsen’s garbage can model, the paper argues that existing organizational structures routed generative AI into integrity and compliance channels while excluding the labor-market signals that would have prompted a readiness response. The governance apparatus consumed the institutional bandwidth, committee time, and administrative attention that readiness would have required. Governance structures, by their success in managing risk, made the harder work of preparation invisible and untimely.
The paper identifies a deeper problem beneath the governance-readiness gap. Institutions treated AI primarily as an integrity threat, building enforcement infrastructure around the question “did the student do this work?” Generative AI has made that question insufficient. When a model can produce the artifact the credential was designed to certify, the assessment model itself requires redesign. The dominant institutional response protects an assessment framework whose validity has changed. The outlier institutions that broke the pattern share a common feature: they started from an external question about what graduates need rather than an internal question about what might go wrong.
The paper reads the outlier case at the variable where the break occurred. The single institution in the dataset that originated its AI strategy from employer input placed its career services unit at the table where the institution set AI policy. The fifty-two other career services units in the dataset have equivalent analytical capacity. What differs at the one case is procedural: the unit has standing in the channel where the institution decides. The paper also reads two independent-discipline findings that converge on the execution-gap reading from outside the governance evidence base. Correspondence with Gene Roche, a former career center director, surfaces a methodological limitation of public-source profiling that recurs as the substantive finding: career services is not reported as an institutional AI function because the field does not think of it as one. A Journal of Macromarketing commentary by Reza Barkhi and Michelle Seref, written from inside business school administration without reference to the governance evidence, reaches the same execution-readiness finding from a different vantage. Three readings converge on one pattern at three evidentiary distances: the cross-institutional coding, the inside informant, and the independent-discipline commentary.
Keywords: generative AI, higher education governance, workforce readiness, institutional isomorphism, assessment, academic integrity, loose coupling
JEL Classification: I23