MIT AI Report: Office Hours and Faculty Trust Erode

MIT AI Report: Office Hours and Faculty Trust Erode

MIT, the university most closely tied to AI's origin story, now has an expert committee warning about what the technology is doing to campus life. Its report, published in June 2026, argues that universities need to rethink their approach to AI from the ground up. The committee describes declining attendance at office hours, fewer study groups, and a weakening of the trust between faculty and students.

The report also points to a preparation gap. A fall 2025 survey by "The Tech," MIT's student newspaper, found that more than two-thirds of students considered AI important for their careers. Only about a quarter felt prepared. The committee wants AI literacy built into introductory courses immediately.

The quiet erosion of campus routines

The committee says students already use AI everywhere, and the side effects are visible. Fewer students come to office hours. Participation in online discussions has dropped. Study groups in dorms and libraries are getting smaller. Faculty, meanwhile, find it increasingly hard to tell what students have actually learned.

The report's guiding principle is "augmentation not automation": AI should extend human abilities, not replace them. Its authors argue that getting a correct answer from a chatbot creates an illusion of learning. Over time, this can lead to intellectual surrender, with students turning to AI as soon as a problem gets hard.

Policing AI is damaging relationships

Faculty report that enforcing rules against unauthorized AI use is straining their relationships with students. The committee explicitly advises against AI text detectors. These tools are unreliable and often flag work by non-native speakers or neurodivergent students as machine-written. They also invite an arms race with "AI humanizers," programs designed to make generated text read as human.

Lockdown browsers, which lock and monitor a computer during an exam, are no better, according to the panel. It calls the current generation buggy and says it feels like surveillance.

Students have their own complaints. They fear false accusations, and they see a double standard when instructors use AI for slides, feedback, or grading while restricting it for students. The committee recommends transparency rules for faculty as well. For theses and dissertations, every paper should disclose its AI use, and AI may never be listed as a co-author.

A flagship program under pressure

One of MIT's signature programs is also exposed. Some faculty members are considering AI agents instead of students as research assistants - a trend tools that turn models into lab assistants make more tempting. That would hit the Undergraduate Research Opportunities Program (UROP), which places undergraduates in faculty research. 93 percent of the class of 2025 took part at least once, and 58 percent of faculty served as mentors. The panel argues that UROP exists to train students, not to supply cheap research labor, so swapping them for AI would defeat its purpose.

Course-by-course rules, not one policy

The committee rejects a single institute-wide rule. A poetry seminar and a course on mathematical proofs relate to AI in very different ways, so one policy would be too loose for some classes and too strict for others. Instead, each course should start from its learning goals, then design assessments, and only then decide which AI uses to allow. The policy and its reasoning should appear in the syllabus.

The report also recommends more oral exams, semester portfolios, in-person discussions, and project-based work.

Access is not equal

Cost is another concern. MIT gives all members access to several AI models through its Parley platform, and faculty and graduate students receive $30 per month in free credits. But premium subscriptions from OpenAI, Google, and Anthropic cost several hundred dollars a month. The committee warns that this gap could translate into real differences in performance.

What other research shows

MIT's findings line up with data from elsewhere:

  • Harvard: About 87.5 percent of respondents in 2024 used AI, nearly half at least every other day. Around 25 percent said it led them to visit office hours less, ask instructors for help less, and skip assigned readings.
  • UK: Usage among full-time students reached 95 percent by the end of 2025, and students from wealthier households used AI more often.
  • Anthropic: Students offloaded higher-order thinking such as analysis and creation to Claude in nearly half of the conversations analyzed.
  • China: A long-term study of more than 26,000 students found AI raised homework grades by 18 percent, but exam scores fell 20 percent after six months.
  • UC Berkeley: Across more than 500,000 grades, the share of A grades in writing- and coding-heavy courses rose by 13 percentage points since ChatGPT launched. The rise was concentrated in homework-heavy courses, which points away from genuine learning gains.
  • Brown: Average scores fell from 96 percent on a take-home exam to 48.6 percent on the in-person follow-up.

One study cuts the other way. At Vrije Universiteit Amsterdam, legal scholar Thibault Schrepel ran a two-year experiment with three randomly assigned groups: no AI, AI without guidance, and AI with training. The no-AI group finished last in both years. Even the unguided AI group did better, although its members often accepted AI suggestions uncritically in class. Schrepel ended up rejecting his own assumption that AI only helps when paired with structured training.

Our Take

The MIT report reads less like an anti-AI manifesto and more like an admission that the old enforcement model has failed. Detectors misfire, lockdown software feels like surveillance, and students notice when faculty play by different rules. Shifting the focus to assessment design is the practical response.

The research picture is mixed, and that matters. Homework gains paired with weaker exam results suggest that much of the measured improvement is AI doing the work. Schrepel's results suggest the tools can still help when the setting is right. The open question is which conditions make the difference.

The access issue also deserves attention. As vendors push capable models behind paid tiers - Google recently limited free Gemini users to a lighter model - the gap between what students can afford may widen. It is worth watching whether other universities adopt MIT's course-level approach, and whether programs like UROP manage to keep humans at the center of research training.