This essay is by Emily Barton, who spent the summer of 2026 researching how universities are actually adopting AI, alongside Kin Lane, for API Evangelist. It is published as the companion to Kin’s essay, Universities Are Building AI Infrastructure Worth Having and Not Publishing the Contract. His half looks at the machine layer. This half looks at the people who have to live with it.
My introduction to Artificial Intelligence was an opportunity to cheat during pre-algebra in middle school. I was struggling with my homework while a friend next to me was breezing through the packet. When I asked for help, they showed me the app Photomath, which had solved all the problems for them step-by-step. It seemed guaranteed to get you an A, but could the teacher catch you using it? Do they even know the app exists? The relationship between me as a student, AI, and my teachers has only become more complex since then. As I embarked on researching AI in higher education alongside Kin Lane for API Evangelist, I considered the perspectives of all the players and found a tense, and evolving discussion. The results of which will shape the future of education.
How did we get here?
As adoption of Artificial Intelligence ripples through most sectors of industry and daily life, it is increasingly the focus of discourse about the future of work, its job market impact, and role in higher education. As a college student currently engaged in the latter, I’ve had a front row seat to the rapid influx of generative AI into academia and its unprecedented disruption of the classroom dynamic. As university administrators and tech giants draft contracts to leverage the power of AI, supervising its usage is often left to course instructors. Professors have been pushed to the front lines of redefining academic integrity on their syllabi, policing AI usage in their classrooms, and navigating labor shifts that strain relationships with students, unions, and leadership.
Most of America’s universities including leading institutions (such as University of California Berkeley, Stanford University, Cornell University, Harvard University…) have a default stance on the use of AI, which is assumed to be prohibited unless explicitly stated otherwise in the course syllabus. Professors thus become the primary boundary setter for ethical and practical AI use in higher education. The resulting patchwork of policies that students have to navigate is illustrated in syllabi that have ballooned from a roadmap of learning objectives to an often ad-hoc AI policy handbook.
Full prohibition is a clear policy. It is also a negligent approach to education, given that AI will absolutely be part of many future jobs. In the context of a competitive entry job market, students may feel pressure to do whatever it takes to stand out to future employers. Thus, when making these policies, professors must keep in mind that students will push the limit of what is allowable.
Indeed, AI has only exacerbated a normalization of cheating that has been growing in recent years. Take, for example, the Stanford Honor Code, written by students in 1921, which dictates that students will report peers they suspect of cheating. The assumption is that cheating is exceptional behavior rather than engaged en masse. However, the Honor Code violations tripled between the academic years 2018-19 and 2020-2021, while student reported violations dwindled from a few to none. Following the results of a pilot program, students and staff voted to allow proctoring in all Stanford classes. For the first time in over 100 years, student reporting is no longer considered sufficient. The advent of GenAI has pushed the limits of student accountability, and the result is increased instructor intervention.
This principle of students being empowered to create their own classroom guidelines can still be adapted for AI. At a small private liberal arts college in Wisconsin, Beloit College’s Tamara Ketabgian is doing just that, encouraging her students’ critical thinking about their use of AI in the classroom by having them draft their own AI Code of Conduct. Students concluded that they would “use A.I. to survey existing scholarly materials on our course topics, but that they would not use A.I. to generate specific text used in their written assignments.” Having the students reflect and define these norms for themselves, as a collective critical thinking exercise, has the potential for higher rates of adherence.
Professors Become the AI Police
Creating this type of policy is only the first step. Enforcing it is an ongoing arms race of detection methodologies and models devised to evade them. Schools regularly advise against AI detection tools due to concerns of accuracy, bias, and student privacy (FERPA) protections. Beyond the shortcomings of detection technology, there is also the student-professor relationship that deteriorates when all assignments are run through AI detectors. This assumes guilt until proven otherwise which is not a basis for a trusting relationship.
Okay, so, no AI detectors. This requires professors to spend time evaluating a student’s demonstration of course material mastery the old fashioned way, based on familiarity with individual students and the idiosyncrasies of a growing number of AI generators. However, the classic methods for handling academic dishonesty can’t keep up with the normalization of AI use and administrators who are slow to make changes.
A case in point came earlier this year when Brown University’s Roberto Serrano decided to make his Welfare Economics and Social Choice Theory class midterm a take-home test, and enrollment almost tripled (from 30 to 86 students). Historically, the average score for this midterm was 65-80%, but as a take-home test the average score jumped to 96%. Feeling something was “fishy”, Serrano ran the test through ChatGPT and got similar responses to his high-scoring students. He told his class that if the in-person final has a similar score distribution to the mid-term that he would count the midterm grades; otherwise he’d assume a gross amount of GenAI usage and discount the mid-term. As a result, 18 students dropped out, nine didn’t take the final, and 19 failed the class. The average score was a historic low of 48.6%.
Serrano reported this data and his suspicions to Brown’s Standing Committee on the Academic Code but only got a response when he later went public with the story. When his department chair did reach out, it was to request he submit a complaint for each student with evidence of their cheating. Serrano considered completing 86 individual submissions to be “ridiculous” as he is left unsupported in this unprecedented case of cheating.
University academic dishonesty mechanisms have assumed plagiarism is a rarity, making streamlining unnecessary. In the era of GenAI, this Brown example demonstrates how burdensome on professors it is when AI use is as widespread as the majority of an undergrad class. Furthermore, instructors are not compensated for taking on this extra work, requiring them to weigh their morals with all the other demands on their time when it comes to preventing or reporting cheating. A dedicated academic integrity department is one approach that would support instructors while sending a clear message to students that cheating must not be normalized.
Communication Breakdown; The Path Forward
AI usage and the ideological questions that arise from it, require patience and communication to resolve. This discussion needs to be between staff, instructors, administration, and students because without cooperation there is distrust and upset. Today we see this play out in the nation’s largest public university system, the California State University system (CSU). In February of last year, faculty and staff across the CSU system received an email breaking the news that CSU was partnering OpenAI to become “the nation’s first and largest AI-powered public university system.” This was breaking news to all involved except the two professors and one student on an unpublicized committee, who were consulted. None of Cal State’s 27,000 faculty members or 460,000 students had heard anything about the institution’s “AI-Empowered CSU” initiative before they opened their email.
This lack of administrator and instructor consultation on AI implementation is a recurring pattern. According to an American Association of University Professors (AAUP) survey of members across two hundred campuses, 71% reported AI initiatives are led by administrators who then push to implement policy into teaching and research without meaningful input from faculty or students. Staff and faculty are upset about the implications of this partnership on their teaching, while also experiencing a sense of betrayal by the administrators who went out of their way to keep this development from those who will be affected by it most: instructors and students.
CSU’s approach has energized a response to this perceived breach of trust and concerns over AI’s notoriety “…for its disruptive impacts on teaching”. Over the past year, Cal State faculty have rolled out petitions, organized public events, media commentaries, and are circulating an open letter to Mildred García, the Chancellor of the CSU System, calling for her to cancel the contract with OpenAI and to use the savings to protect jobs at CSU campuses facing layoffs. The California Faculty Association is so concerned about this that, in conjunction with AAUP, they filed an unfair practice charge against CSU management for “failing to meet and confer about the impacts of the AI initiative on our working conditions”. This charge has not yet been adjudicated, but the filing alone is a statement of the importance of these AI contracts to educators and that judicial oversight may become another player in these discussions.
The irony is that these entities want the same thing. Clear and consistent communication between everyone would go a long way in ensuring a quality education for their students. The AAUP suggests that shared governance policies create an environment more conducive to incorporating the perspectives and expertise of those at every level of university operations. Under such policies it would be the shared responsibility of faculty, administrations, and governing boards to govern the university. The weight of each group’s voice on a particular issue determined by their proximity to and expertise on that issue. Regardless of the framework used by colleges or universities, AI is challenging these systems and relationships, changing them in ways we do not yet fully know.
This is the first generation of students to navigate having AI in their education and in the workforce they hope to enter. It is also the first generation of professors expected to guide students through this complex new system. Having students be the focus of this conversation and policies is reasonable. However, without also considering how AI usage is enforced by those on the front lines does a disservice to everyone. Professors are not adequately supported with their compounding responsibilities, which risks having policies inconsistently upheld, creating further ambiguity for students. Thus, this threatens to create a generation of workers who appear equally qualified despite being of unequal skill.
What is the solution? A monolithic rule, or are we forever going to be in a project-to-project AI policy limbo, and who decides that? There can never be a single, static answer. There must be policies and enforcement systems in an evolving compromise through the empathetic collaboration between university administration, staff, students, and AI developers.
Emily Barton researched how universities across California are adopting AI during the summer of 2026 for API Evangelist. You can connect with her on LinkedIn. Her research into the California system, including UC Irvine’s ZotGPT platform, is what Kin Lane’s companion essay is built on — read it here: Universities Are Building AI Infrastructure Worth Having and Not Publishing the Contract.
