Over the last few months, the University has made headlines for implementing a wide array of approaches to artificial intelligence.
In June, University President Paul Alivisatos announced that UChicago would partner with AI company Anthropic to offer students free access to their flagship large language model (LLM), Claude, ahead of the 2026–27 academic year. Not long after, the Law School released a detailed strategy statement outlining a nuanced approach to AI, banning its use in introductory sequences while slowly incorporating it in upper-level coursework. And just last month, the Maroon reported that social sciences (sosc) Core classes would ban the use of technology in class and AI altogether across all sections.
Despite (or perhaps because of) the diversity of approaches, UChicago continues to lack a University-wide AI policy, instead leaving matters of regulation, detection, and enforcement up to programs, like the social science Core, or individual professors. Peer institutions vary in their approaches, with several opting to implement university-wide policies.
“It is absolutely not a top-down thing,” English department chair Benjamin Morgan said, describing how AI policies are developed for the classroom. “There is no kind of centralized directive coming from above saying, ‘Here are the policies that you must implement.’”
The lack of a centralized AI policy comes with both benefits and drawbacks. “There being a universal policy would make it easier to do [disciplinary] casework and make it easier for students to know what is and is not accepted,” said Jessie Wang (A.B. ’26), who over the last two years served as president of the Student Advocate’s Office (SAO), which assists students charged by the University with disciplinary infractions. On the other hand, “with the differences between the humanities and STEM classes, the standards for what can and can’t be done differ a lot.”
Another obstacle to a University-wide standard is the traditional UChicago belief that “professors should have full control over the way that they teach in their classrooms,” Wang said. “A universal AI policy would definitely get in the way of that for some people.”
Perhaps the Law School’s recent strategy statement can offer clues to the University’s developing philosophy regarding AI. In the coming academic year, AI will be banned from all introductory and required courses, but gradually incorporated into higher-level electives, several of which deal with AI directly. The Law School’s policy centers on the cultivation of “essential human” skills, such as judgment, reasoning, and critical thinking, so that students are able to use AI responsibly in the long run.
The University as a whole, meanwhile, is in what can only be described as an experimental phase. As early as 2025, Alivisatos announced the establishment of AI working groups to develop frameworks for the University’s approaches to AI both in the classroom and in research and scholarship. Nearly two years later, the University remains in that exploratory stage, allowing departments and professors space to work out what makes sense for their own disciplines. Whether that will—or should—change is another question entirely.
How have AI policies shaped up thus far?
Lisa Rosen, executive director of UChicago’s Science of Learning Center and a professor in the Committee on Education, spoke to the Maroon about how her policies have evolved in the past few years. She previously allowed students to use AI on their take-home essays as long as they disclosed it, and students were asked to submit their papers as a Google Doc with version history accessible. However, Rosen noticed that this combination did not necessarily lead to honest reporting of AI use.
She has since “redesigned” the course so students must choose between an oral exam or an in-class written exam, making it “more or less impossible for them to cheat.”
Rui Zhao, who teaches ECON 100: Principles of Microeconomics, one of the largest classes on campus, has changed the structure of her exams in recent years to focus less on factual knowledge and computation and more on explaining what calculations mean and how they are derived, in part as a method of combatting AI. However, she doesn’t unilaterally ban AI use for class assignments. “Even if I tell you [that] you shouldn’t use it, it just puts you in a disadvantageous position,” she said. But homework itself is part of the training process. She believes only once students internalize the trade-off can they understand they are “cheating themselves” by using AI.
In addition to determining AI policy, professors are also responsible for identifying suspected misconduct. But the absence of a common AI standard makes both detection and disciplinary action more difficult.
“What cheating means in one class can be different from what cheating means in another class,” Wang, the former SAO president, said. A student who uses AI to edit prose in an English course or debug code in a computer science (CS) course may be violating the rules of one course while acting within them in another. For that reason, the most important element in any AI-related disciplinary case is the syllabus of that particular course, Wang said.
The stakes of the matter have risen quickly. According to Wang, the SAO had “basically no AI cases” two years ago. Today, she estimated, “maybe half of our cases or more are AI-related.” According to the University’s disciplinary action reports, there were four, three, and two AI-related infractions in the last three academic years, respectively, with punishments including formal warnings, disciplinary probation, and suspensions up to three quarters long. Data on the 2025–26 academic year has yet to be released.
Recognizing the challenges of detecting and proving AI misuse, the Office of College Community Standards (OCCS) considers a range of evidence in such cases, from AI detectors to Google Docs version histories, though even its standards are continually evolving. According to Wang, the OCCS did not give AI detectors much weight due to concerns about their reliability, up until this past academic year. Now, however, it uses Pangram, a paid detection software that a recent study conducted by the Booth School of Business found was “the only AI detector maintaining policy-grade levels on our main metrics.”
Deteriorating student–professor relationships
For many professors, the larger concern when it comes to AI detection policies is their negative effect on student-teacher relationships. Morgan, chair of the English department, argued that the intense focus on detection and discipline has had the unintended effect of turning professors into police, encouraging them to approach students’ work with suspicion. “I’ve heard many times from instructors that this feeling—that you’re never sure whether something a student has turned in is their work or AI work—has really damaged the relationship between students and instructors,” he said. Cognitive science undergraduate program coordinator Melinh Lai moved away from a policing model precisely because she wanted to avoid such problems. “I did not want to be in this place of trying to figure out who is using AI and how I should punish them for using AI,” she said. “It felt like the relationship I was forming with my students from that perspective was going to become very adversarial.”
Rosen went even further, arguing that the ease with which students can conceal or lie about AI use “poisons the well” of the classroom, disrupting “the foundation of trust on which learning rests.”
She continued, “If you want to learn from someone, you have to trust them. You have to be willing to say what you don’t know. You have to be willing to say, ‘I’m confused. Can you help me? Can you clarify this for me?’ If that trust isn’t there, learning becomes really difficult.”
How has AI interfered with learning outcomes?
It is hard to pin down whether AI’s increasing prevalence in the classroom will be a net benefit or detriment to learning. Professors are asking the same questions.
CS professor Nick Feamster believes the classroom is a simulator for the real world, and chooses to “acknowledge the environment that we are living and operating in.” Although the CS major has declined in popularity as AI has become better at programming, he believes that AI can make the computer science degree more popular by changing its “grungy and inaccessible” reputation. “[Computer science] has never been about just writing code,” he said. Instead, it emphasizes more universal skills like reasoning, abstraction, and testing through the study of algorithms and data structures.
Sanjay Krishnan, an associate professor of CS, was happy to note that during office hours last academic year, students were less “obsessed” about completing homework assignments and more interested in discussing the course material—something he attributed to AI becoming more common. However, he noticed a limited understanding of class content while correcting midterms. He pointed to three particular questions in his midterm that students struggled to answer, which in previous years students typically got “100 percent correct.”
The negative effects of AI on learning outcomes are not limited to UChicago. After suspecting that his students had cheated with AI on a take-home midterm due to unusually high scores, Brown University economics professor Roberto Serrano announced that the final exam would be in-person and strictly monitored. Following the announcement, 18 students dropped the course, nine skipped the final, and 19 failed. He compared the scores of the midterm to the final and noticed a concerning drop in the latter.