Here is my work. Here is evidence of how I was involved in producing it. I am willing to stand behind it.
That’s what a student should be able to say. Today doing so generally means exposing more of the process than should be necessary.
Many of the fixes higher ed has reached for make the relationship between students and instructors worse, not better. Detectors mislabel honest writing. Lockdown browsers box students in and call it trust. Policing has replaced trust in ways that are unhealthy for both students and instructors. None of it answers the question that actually matters: was the student involved in producing their work?
MIT’s ad hoc committee spent much of this year on these questions and published its final report on August 13, 2026. It runs 38 pages, and most people would be tempted to read the summary. What follows are the details that a summary leaves out.
The report begins from a philosophical point of view (2.4).
Leaning into learning means creating a new “social contract” between teachers and students. All of us who teach at MIT will need to be prepared to help students understand both that the process of education is necessarily a productive struggle, and that the most important product of their education is not a GPA or a diploma but themselves: their personal growth and intellectual maturity and the development of their own imagination, insight, and judgment.
We need to help them develop metacognitive abilities to think about thinking, to engage in reflective practices, and to enhance their sense of personal agency.
To me, that’s an argument to protect the conditions in which learning occurs, not to optimize primarily for control and inspection.
That philosophy collides with how detection actually works (3.1.9).
Nevertheless, we recommend against relying on AI detectors, as it risks an arms race in which students respond to automated detection by using increasingly powerful “AI humanizers” to remove signals that AI detectors are cued to catch. The result: a lot of effort on both sides that in the end serves no one.
More generally, stepping up “policing” around AI use builds an adversarial atmosphere of distrust between instructors and students, which understandably hurts students’ motivation and morale.
...so-called “lockdown” browsers for conducting evaluations of students. These are online testing environments that take over the computer during an exam, preventing access to any online resources outside of those provided as a part of the test. The committee recommends that MIT study such tools, but notes that the current generation is buggy, error-prone and feels like surveillance.
The costs aren’t limited to detection tools either (3.1.2).
Already, many are increasing the weight of exams in their grading or asking students to write or code during class time. However, these tactical solutions come with a cost: For instance, overemphasizing in-class evaluations means reducing students’ incentive to invest themselves in the difficult, time-intensive p-sets and projects it takes to build the ladder to mastery.
By definition, shifting assessments to time-limited class periods reduces how much thought and deliberation students can put in. If we want students to care about and know how to create and recognize worthy work – work of scope, rigor, creativity and thoughtfulness – quick, high-stakes evaluations embody the opposite of the signal we want to convey to them right now.
As I see it, institutions need confidence that a student was involved in producing the work. But moving more work into controlled environments can reduce the very deliberation and productive struggle institutions want students to experience. Attempts to recover that confidence by judging the finished artifact have added another layer of suspicion to the student-instructor relationship.
The MIT report says the evidentiary standard is unsettled, and that its Committee on Discipline (COD), the body that handles integrity cases, does not treat AI detector output alone as sufficient (3.1.9).
MIT needs to make sure that Institute policies clarify what evidence would be required to bring an academic integrity case forward when AI is involved; the COD itself does not consider AI detector output alone sufficient.
I think those recommendations point in an important direction: away from relying on detection and toward evidence of how the work got done (3.1.9).
Other technical tools can also be helpful. For example, instructors can require students to do their work on platforms that capture a history of versions, and to submit the history along with their work. This can provide useful process evidence – for example if a student were to submit an assignment within a few minutes though comparable work would typically take hours.
And students themselves often find these tools useful for reflecting on how their ideas have evolved.
That shift in emphasis reduces the evidentiary question, in my view, to something more primitive: not what can be inferred after submission, but what can be observed about how involved a student was in producing the work.
When it comes to provenance for problem sets, projects and outside-of-classroom exams, I think there’s a more primitive question worth asking first: what evidence exists of the student’s involvement in producing the work? The reason it is primitive is because this question does not require judging the words or deciding whether AI was used appropriately. The question also does not attempt to get into authorship, learning, originality, honesty, or mastery. That narrowness is deliberate, not a shortfall. It is what keeps the evidence from becoming another form of the surveillance MIT is warning against.
This question is valuable because the evidence is direct: it comes from the production of the work itself, not an inference made afterward from the artifact.
The student’s position changes too. Their focus moves off the final artifact, which they know matters to their grade, and onto carrying evidence of their own involvement. That’s what a student can’t yet offer today.
The same signal that answers the question for the instructor also helps students to produce affirmative evidence of involvement, rather than only defend themselves against a downstream inference. Process evidence also has to stay within an institution’s privacy policy. It can come from reading the work, or it can come from somewhere else. Version history points in the right direction because it moves the evidence upstream, toward the production of the work. But it still exposes the content of that process. I don’t think that exposure should be necessary. Student-friendly process evidence, as I would define it, should answer a narrower question: was the student observably involved in producing this work? Answering that should not require reading the words they wrote along the way. This approach gives students agency because it focuses on their activity without collecting information to pass judgment.
MIT’s report describes a problem far larger than anything one product can address. But here’s what I think is solvable right now: giving students agency over authorship, a way to show their involvement in written work without exposing the words themselves.
The student stops defending against an inference and starts offering a signal: evidence that they were involved in producing the work.