At Harvard, the AI Rules Depend on Your Instructor. The Extension School's Policy Says So in Writing

Students in Harvard Extension School courses can face very different rules on artificial intelligence depending on the class, and the school's own policy is designed that way.

What the policy says

The Extension School's academic integrity page says its policy "prohibits students from representing work as their own that they did not write, code, or create" and that "submission of computer-generated text without attribution is not allowed," naming tools such as ChatGPT.

It then gives instructors wide latitude: "Given the wide range of learning goals in courses at the Extension School, individual instructors may create course-specific policies that differ from and supersede these guidelines." That may include "forbidding any use of generative AI, even if only to improve the readability or language of your submission." If an instructor allows AI for an assignment, students must still acknowledge its use.

The school is also piloting Turnitin, a text-matching platform, in select courses at instructors' discretion, and says faculty review its reports rather than treating matches as proof of misconduct.

How instructors are using that freedom

With most Extension School courses online, the decisions have largely fallen to individual instructors, and the results range from bans to structured use, The Harvard Crimson reported.

In Bios E-10, "Introduction to Biochemistry," instructors Robin Haynes and Roopali Roy prohibit AI on graded work, the Crimson reported. Their problem sets are "more conceptual in nature," they wrote, and AI could easily "draft a comprehensive answer that would probably earn them points." Students may use AI to build study guides or practice problems. "AI is not accessible during the exams and by using AI on the problem sets, it will only hinder their ability to think through problems on the exams," they wrote.

In Math E-3, "Quantitative Reasoning," instructor Graeme D. Bird dropped an outright ban after talking with colleagues and now allows AI only to explain confusing terms and generate practice problems. "I have no idea how many of them actually do use it though," he told the Crimson.

The Extension School's Writing Program goes furthest. Its policy says, "AI technologies may not be used in any phase of the process, including brainstorming, outlining, drafting, revising, and editing," according to the Crimson.

Testing in an online school

Bird relies on closed-book, human-proctored final exams, and Haynes and Roy added in-class pop quizzes where students can't use AI, the Crimson reported. Haynes and Roy said recent exam averages are close to scores from five years ago, and Bird said graders see only "rare" instances of blatant AI use. "I feel like this is a work in progress," Bird said.

Training the people who teach sections

On the main campus, the Faculty of Arts and Sciences now requires all teaching fellows, teaching assistants and course assistants to complete training at the Derek Bok Center for Teaching and Learning, the Crimson reported in a separate story. The center trained 1,556 section leaders from late August through early September, faculty director Karen L. Thornber said, more than double the roughly 700 first-time TFs and TAs trained last fall.

"I realized a couple things — first, just a small percent of the TFs, TAs, and CAs were being trained, and secondly, that the training really didn't meet current needs," Thornber said. The center piloted the program with GenEd TFs in January 2025; FAS Dean Hopi E. Hoekstra made it mandatory for new TFs and TAs in fall 2025, and this fall it expanded to undergraduate course assistants and returning staff.

Part of the training covers teaching controversial topics. Not everyone found it useful: a first-time TF for Ec 10 said the participation-focused strategies didn't fit a large lecture course, and a Math 22a TF called it a "box to tick." Thornber said the center will bring in more STEM instructors next year. "We weren't doing enough explicitly to help our STEM colleagues," she said.

The common thread

Both stories describe Harvard handing classroom decisions to the people closest to the course, whether on AI or on how sections are run, while building more central support around them. The Crimson reported that many Extension faculty credit the school with useful workshops and forums on classroom technology even as it keeps a decentralized stance on AI.

What is not known

Neither report said whether the Extension School plans a school-wide AI rule, or how many courses ban AI outright.