An MIT committee calls for rethinking aims, assessment and the in-person community in the face of an AI that can already do almost any written work
The ad hoc committee of the Massachusetts Institute of Technology proposes eight principles and a set of recommendations for adapting university teaching to generative AI.
The document is the final report of the Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training at the Massachusetts Institute of Technology (MIT). The institution’s leadership commissioned it in January 2026 to assess current use of generative AI, identify innovations in teaching and assessment, and propose a policy on its use. It is dated 13 August 2026 and is co-chaired by Eric Klopfer and Sam Madden; the committee brings together undergraduate and graduate students, faculty from all of the schools and staff from units such as the libraries and the Teaching and Learning Lab. It is an institutional document, addressed to the MIT community.
The committee concludes that the issue goes deeper than drafting a rule: it affects the structure, purpose and value of a university education. According to the report, AI can give reasonable answers to almost any written task in the undergraduate curriculum, and it warns: “This is not a moment for patches and duct tape.” Among the data it provides, in the autumn 2025 technology survey more than two thirds of the students who responded believed that AI would matter in their careers, but only 25% felt that MIT was adequately preparing them to use it.
The eight principles are: be humble, be bold, put people at the centre, bet on learning, teach with intention, recognise that there is no single solution, augment rather than automate, and think beyond the classroom and the campus. The fourth speaks of productive effort and a new social contract between teachers and students; the seventh asks for AI to be used, in the report’s words, “to augment and enhance curiosity, creativity, and learning, not automate them”. The fifth proposes defining the purpose of learning first, rather than asking whether AI should be allowed or banned.
The recommendations are grouped into three blocks. The first, on educational processes, calls for a review of the aims of each subject, the use of oral exams, semester portfolios or out-of-class work accompanied by discussion in class, and stronger project-based learning and in-person social learning. It also calls for a justified AI policy in every course, in a standard format with clear and consistent guidelines. It advises against relying on AI detectors, which it considers unreliable, and describes current lockdown browsers as flawed and close to surveillance, although it recommends studying them; it prefers version history and staged feedback. The second block, on community and the residential experience, recommends that faculty disclose their own use of AI, that effective, responsible and ethical use be taught from students’ induction sessions onwards, and that every thesis include a statement on how AI was used, without it ever being listed as a co-author.
The third block proposes support structures: a standing committee on AI and education, AI leads in each school or department, fellows and an implementation team, a fund for pilots, ongoing training and support for faculty, and metrics on use, participation and outcomes. It also includes equitable access to technology, a platform that does not tie the community to a single model provider, and monitoring of costs and environmental impact. These measures depend on the resources of a particular university.
This report is thematically related to other documents in the library: assessment in the age of AI, such as Responsible Assessment in the AI Era, or academic integrity, such as A Comprehensive Academic Integrity (CAI) Framework: An Overview. The Diario has not compared their content with that of this report. For a teacher, the most useful parts are the transferable ideas: start from the learning objectives before deciding what is permitted, state in each course which uses of AI are acceptable and why, and diversify assessment with oral or process-based components. It is best to begin with the block on educational processes, the one closest to the classroom. As a limit, the context is that of a particular research university, so applying it to other levels or schools requires adaptation.
An MIT committee calls for rethinking aims, assessment and the in-person community in the face of an AI that can already do almost any written work
The ad hoc committee of the Massachusetts Institute of Technology proposes eight principles and a set of recommendations for adapting university teaching to generative AI.
The document is the final report of the Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training at the Massachusetts Institute of Technology (MIT). The institution’s leadership commissioned it in January 2026 to assess current use of generative AI, identify innovations in teaching and assessment, and propose a policy on its use. It is dated 13 August 2026 and is co-chaired by Eric Klopfer and Sam Madden; the committee brings together undergraduate and graduate students, faculty from all of the schools and staff from units such as the libraries and the Teaching and Learning Lab. It is an institutional document, addressed to the MIT community.
The committee concludes that the issue goes deeper than drafting a rule: it affects the structure, purpose and value of a university education. According to the report, AI can give reasonable answers to almost any written task in the undergraduate curriculum, and it warns: “This is not a moment for patches and duct tape.” Among the data it provides, in the autumn 2025 technology survey more than two thirds of the students who responded believed that AI would matter in their careers, but only 25% felt that MIT was adequately preparing them to use it.
The eight principles are: be humble, be bold, put people at the centre, bet on learning, teach with intention, recognise that there is no single solution, augment rather than automate, and think beyond the classroom and the campus. The fourth speaks of productive effort and a new social contract between teachers and students; the seventh asks for AI to be used, in the report’s words, “to augment and enhance curiosity, creativity, and learning, not automate them”. The fifth proposes defining the purpose of learning first, rather than asking whether AI should be allowed or banned.
The recommendations are grouped into three blocks. The first, on educational processes, calls for a review of the aims of each subject, the use of oral exams, semester portfolios or out-of-class work accompanied by discussion in class, and stronger project-based learning and in-person social learning. It also calls for a justified AI policy in every course, in a standard format with clear and consistent guidelines. It advises against relying on AI detectors, which it considers unreliable, and describes current lockdown browsers as flawed and close to surveillance, although it recommends studying them; it prefers version history and staged feedback. The second block, on community and the residential experience, recommends that faculty disclose their own use of AI, that effective, responsible and ethical use be taught from students’ induction sessions onwards, and that every thesis include a statement on how AI was used, without it ever being listed as a co-author.
The third block proposes support structures: a standing committee on AI and education, AI leads in each school or department, fellows and an implementation team, a fund for pilots, ongoing training and support for faculty, and metrics on use, participation and outcomes. It also includes equitable access to technology, a platform that does not tie the community to a single model provider, and monitoring of costs and environmental impact. These measures depend on the resources of a particular university.
This report is thematically related to other documents in the library: assessment in the age of AI, such as Responsible Assessment in the AI Era, or academic integrity, such as A Comprehensive Academic Integrity (CAI) Framework: An Overview. The Diario has not compared their content with that of this report. For a teacher, the most useful parts are the transferable ideas: start from the learning objectives before deciding what is permitted, state in each course which uses of AI are acceptable and why, and diversify assessment with oral or process-based components. It is best to begin with the block on educational processes, the one closest to the classroom. As a limit, the context is that of a particular research university, so applying it to other levels or schools requires adaptation.
