MIT AI report takes first step towards an Institute-wide response
Klopfer and Madden: “As we consider any kind of changes, they must be made alongside the students who are impacted”
On Aug. 13, 2026, MIT’s Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training released their final report after nine months of deliberation. Charged by the Institute in January, CMS-W Professor Eric Klopfer and Faculty Head of Computer Science Sam Madden directed the committee of staff, students, and professors, many of whom had previously studied at MIT.
The committee, as established in a letter from MIT leadership, sought to assess AI usage on campus, find opportunities to leverage AI in education, and, ultimately, propose an AI use policy.
Since the first public release of ChatGPT in late 2022, the widespread adoption of generative AI has forced universities to react in varying ways.
For example, Harvard requires each instructor to clearly define course rules around AI use without establishing campus-wide rules. Brown, based on its own report on AI in teaching and learning, opted for a broadly applicable set of guidelines for generative AI while allowing departments and programs to develop additional standards for their specific areas.
Other institutions have leaned into the new opportunities that AI presents. Purdue has expanded its curriculum with an AI course catalog and has designed multiple new undergraduate majors in AI, similar to MIT’s Artificial Intelligence and Decision Making (Course 6-4) major. Purdue also became the first university to add an AI competency graduation requirement.
To orient its own response to the prevalent use of AI on campus, the committee relied in part on two community surveys in Fall 2025 and Spring 2026. Unsurprisingly, the surveys revealed that most students, faculty, and staff use AI on a daily basis for a variety of tasks, believing that it makes them more efficient. Nearly 80% believed that AI skills would be necessary for their future careers.
But many survey takers also expressed significant anxieties. Some had concerns about the accuracy of models and their impacts on society, intellectual property, and the environment. Among all respondents to the 2026 survey, 48% worried about the cognitive consequences of relying on AI. Only 25% of undergraduates felt that MIT was preparing them to use AI professionally. And in its assessment of the current educational landscape, the committee highlighted decreased engagement among peers, weakened teacher-student relationships, increased difficulty in evaluating student work, and less “creative friction” that drives learning.
Adapting educational processes
“AI can produce credible solutions and provide reasonable responses to almost any written assignment in our undergraduate curriculum,” the committee wrote. The report noted a drop in the number of students spending time in-person learning spaces such as office hours, class discussion boards, libraries, and group study sessions.
To adapt to these realities, the report made a number of suggestions. Instructors should shift their focus toward course content that cannot be easily completed by AI, such as in-class discussions, oral exams, and long-term projects. Coursework should be designed around social learning, such as supervised problem-solving or group projects, which would require more hands-on labor from instructors and TAs. Resources allocated to UROP should expand, and MIT should actively explore other sponsored extracurricular learning, such as the co-op program currently being explored by another committee.
To eliminate confusion about what kinds of AI assistance are acceptable, the committee suggested that every class have a clear generative AI policy. Acknowledging both the need for clarity and the impossibility of a “one-size-fits-all” policy, they advised MIT to create a standardized “menu” of use policies. An appendix outlined four examples, each designed for courses that would permit, limit, require, or forbid AI tools, respectively.
In a statement to The Tech, Klopfer and Madden said the committee has “encouraged faculty and instructors to use one of [their] standard formats,” and that formats will be discussed in fall workshops and forums as well as at the September faculty meeting.
“We expect that departments and instructors will maintain authority over specific policies for a subject, and also believe that there are likely some classes that don’t fit neatly into the template,” Klopper and Madden said.
Although it does not provide concrete solutions, the report also emphasized that instructors must think carefully about how to identify instances of AI cheating. The committee cautioned against unreliable AI detectors, even pointing out how investing in automated policing “risks an arms race” between students and instructors, eroding trust. It also stressed that MIT needs to clarify what kinds of evidence are grounds for referral to the Committee on Discipline.
The current grading system also went under the magnifying glass. “We also talked a lot about grading. We do not have a specific proposal on this,” the co-chairs wrote. But the report acknowledged that grades may become less meaningful, both as metrics for learning and as signals to employers. It also points out that at a gradeless MIT, there would be much less incentive to use AI to cheat.
Implementing AI GIRs is not mentioned in the report. However, the co-chairs did meet with the Task Force on the Undergraduate Academic Program. “We tried to restrict our scope to AI-related changes, but that isn’t always a clear boundary,” Klopfer and Madden said. They acknowledged that changing learning standards in various majors could require students to take courses that teach them about AI, but, as in much of the report, they recommended exploration, not prescription.
“While we see GIRs as a possibility, we don’t see that as the first order of business,” they wrote.
Considering the student experience
Citing a general dependency on AI and an increasing worry about career prospects and purpose in an “AI-saturated world,” the report argued that campus life has become more fragmented and isolated. To reverse these trends, the report argued that MIT needs to invest in what it calls “residential education” — learning that students do together outside of the classroom. Residential education requires trust among students and faculty, and a culture which emphasizes community. To this end, the committee recommended that MIT expand programs like its first-year learning communities and MIT Reads, and to find ways to emphasize to students the importance of relying on themselves and each other in their education.
Although AI is widely available, the report noted that different degrees of access to AI tools could lead to disparities in performance among students. Many commercial large language models (LLMs) provide plans with enormous token counts that cost upwards of $200 per month, a price which is out of reach of many students. The committee recommended continued use of a model-agnostic system, like Parley, and called on MIT to consider expanding the current Parley budget or to provide other AI tools for students as needed. It also pointed out the necessity of a responsible privacy policy for MIT-managed AI tools.
The report also advised transparency about how AI is used to prepare materials and evaluate student work. According to the committee’s survey, many instructors use AI tools for administrative tasks (49%), to create assessments like exams or psets (38%), and to generate lecture notes (36%). While less than 10% of instructors report using AI for teaching assistance or grading, students are still anxious about these use cases. In The Tech’s survey, more than two-thirds were very or somewhat uncomfortable with learning tools like automated teaching assistants or oral exams.
Some graduate students at MIT have particular concerns about the use of AI in teaching and research. The MIT Graduate Student Union (GSU) began contract negotiations with MIT in April to extend its previous contract, which expired at the end of May. In its recent proposals, the GSU has included language that would prevent MIT from using AI to replace or substitute graduate student labor or to monitor or evaluate graduate student employees. The GSU also proposed that MIT obtain consent before using data collected by graduate students for training or developing AI models. MIT has removed this language in its counterproposals.
Daniel Shen, the Campus Chief Steward of the GSU, spoke to The Tech about these issues. “Our understanding was that MIT felt that the union’s initial proposal on this would have been far too restrictive,” he said. The union is not opposed to the use of AI to enhance efficiency in teaching or research, but it believes that “MIT can’t have the right to unilaterally impose these changes on the graduate population without further discussion.”
Although he was not aware of specific cases at MIT, Shen said that the replacement of human instruction with AI was a broad concern in higher education policy, and for the union in particular. The union is also worried about the use of AI tools for surveillance and monitoring. Shen noted the recording and identification of student protesters at many schools as especially worrying.
Future work
The report is only a first step. Its recommendations will need to be discussed and implemented by the entire MIT community. In the committee’s words, “there is a lot to do.”
Highlighting the need for rapid implementation and experimentation, it advised MIT to establish an ongoing committee to continue to investigate the impacts of AI and advise the institute on how to update its policies. They recommended appointing AI leads across schools or even departments to coordinate local decisions about AI, establishing an AI implementation team to help instructors adapt their courses, and creating a pilot fund to provide the resources for training and research on AI in education. They also suggested that MIT should track AI use metrics and closely monitor the costs and environmental impacts of its usage.
While the report only contains a set of recommendations, Klopfer, Madden, and other committee members remain engaged in the process to begin implementing what they can.