Harvard Faculty Say AI Harms Courses as Detection Proves Elusive

Nearly two-thirds of Harvard Faculty of Arts and Sciences professors now say artificial intelligence has harmed their courses, a sharp increase that reflects growing frustration with generative AI tools nearly four years after ChatGPT's public debut.

Sixty-four percent of faculty who responded to The Harvard Crimson's annual survey said AI had a "somewhat negative" or "very negative" effect on their courses this year, up from 42 percent in the previous survey. The findings, first reported by The Crimson, come from responses collected between April 29 and May 22, 2026, from 463 faculty members across Harvard's academic divisions.

Policies proliferate but problems persist

Faculty have responded by tightening course-level AI policies. Only 3.5 percent of respondents said they have no explicit AI policy, down from 10 percent last year. The share of professors who entirely permit AI use dropped by half, from 8 percent to 4 percent.

Yet nearly 90 percent of respondents said they had received student work they knew or believed was produced using AI. Among faculty who "frequently" encounter AI-generated submissions, 48 percent reported very negative effects on their courses—but only 23 percent entirely prohibit AI use.

Formal enforcement remains limited. Just 12 percent of faculty referred students to Harvard's Honor Council or Administrative Board for unauthorized AI use this year, up slightly from 10 percent last year. Referrals were more common among professors who frequently encounter AI work, with 25 percent making formal reports.

History department Director of Undergraduate Studies Mary D. Lewis told The Crimson earlier this year that AI use is "virtually impossible to prove," explaining faculty reluctance to pursue formal cases.

Creative countermeasures emerge

Some instructors have developed workarounds. Course staff for Government 1759 embedded hidden instructions in a take-home final to catch students who copied questions into AI tools. Others have shifted to in-person blue-book exams or require students to submit Google Docs with version histories.

Policy approaches vary significantly by discipline. In a random sample of 600 fall 2026 classes, more than 50 percent of science and engineering courses allowed AI use, compared with 35 percent of social science classes and 27 percent in arts and humanities.

Why it matters

The survey results reveal a widening gap between institutional AI adoption and faculty experience in the classroom. While Harvard has launched AI chatbot advisors and added required AI modules to first-year writing courses, professors report diminishing confidence in their ability to maintain academic integrity. The challenge extends beyond detection: 64 percent of faculty said they feel somewhat or very confident differentiating student work from AI output, meaning more than a third lack that confidence. As generative AI becomes more sophisticated, universities face mounting pressure to develop enforcement mechanisms that match the scale of the problem—or fundamentally rethink how they assess student learning.

College Dean David J. Deming has encouraged faculty to integrate AI into writing and project-based classes and announced plans to establish a "mission-aligned AI use policy" by the end of the academic year. FAS Dean Hopi Hoekstra appointed Physics professor Christopher Stubbs as her senior adviser on AI, according to FAS spokesperson James M. Chisholm.

The faculty concerns coincide with broader worries about academic rigor. Roughly 70 percent of respondents agreed that Harvard students do not sufficiently prioritize coursework, consistent with a 2025 report finding students spend more effort on extracurriculars than classes. Faculty voted this spring to cap A grades at 20 percent, a measure that will formally take effect next year but many instructors are implementing this fall.

The Crimson distributed the survey to more than 1,400 faculty members and received 463 responses, including 292 complete submissions.