MIT Says Stop Using AI Detectors. Your Syllabus AI Policy Now Names You Too.

Your syllabus AI policy needs a line about you. MIT says instructors should disclose AI use in slides, feedback and grading.

On August 25, 2026, MIT released the final report of its Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training. President Sally Kornbluth sent it to the community the same day in a letter titled AI and education: A watershed moment for MIT, writing that the opportunities and risks generative AI poses for MIT’s model of education and research now constitute a watershed for the Institute and for all of higher education.

Every outlet covering it led with the word watershed. That is the quotable part. It is not the part that changed anybody’s Tuesday.

Buried in the same set of letters is a much smaller sentence, and it is the one that landed in actual inboxes: instructors would receive a message that day pointing to guidance for formulating AI policies for the classes they are teaching this fall.

So the practical instruction is not a vision. It is a writing assignment, due before the term gets away from you. And the report specifies who else that policy covers.

The report closes the enforcement door first

Before it asks instructors for anything, the committee removes the option most faculty have been quietly relying on.

The recommendation is direct. The committee report advises against relying on AI detectors, on the reasoning that it risks an arms race in which students respond to automated detection by using increasingly powerful AI humanizers to remove the signals detectors are cued to catch. The report adds that MIT’s own Committee on Discipline does not consider AI detector output alone sufficient to bring a case.

The more interesting finding is not about the software. It is about what the software was doing to the people using it. In the committee’s account, many instructors said that having to police unauthorized AI use was harming their connection to students, a dynamic the report notes was made worse by the fact that AI detection software is quite unreliable.

Read that as a diagnosis, not a complaint. Detection was never really a technology problem, which is the same conclusion we reached in AI Detection Tools Are Failing. It was a relationship cost that faculty were paying every week, in exchange for a signal that does not hold up.

Which raises the obvious question. If you cannot catch it, and you are not supposed to try, what is actually left?

What is left is a syllabus AI policy, and it is yours to write

The committee’s answer is the syllabus. The recommendation is that instructors, and perhaps departments, should make sure that every MIT subject has a clear policy about the use of generative AI, posted prominently in the syllabus and on the course website.

Note the level it lands at. Not the institution. Not the provost. The subject. That is the same devolution we wrote about in Your AI Committee Is Not Slow, arriving now with a deadline attached. MIT says plainly that it recommends against imposing a one size fits all policy on AI use, while also reporting that students are anxious for clarity: in any given course, they want to feel sure about when, where, and why AI is prohibited, allowed, or required.

That combination is the whole assignment in one sentence. No uniform rule is coming to rescue you, and vagueness is the thing students are complaining about. A policy that says “use AI responsibly” satisfies neither half.

The report is careful to say this does not mean every instructor invents policy from scratch. MIT commits to providing a shared framework with a policy menu, disclosure expectations and accountability standards, and points at departments as the natural unit for coordination. But the menu still has to be chosen from, per subject, by the person who knows the assignment.

The half nobody covered

Section 3.2.3 of the report carries a heading that almost no coverage quoted. It reads: encourage instructor disclosure around their own AI use.

The recommendation itself is unambiguous. If instructors plan to present students with content substantially generated by AI, or to use AI for some aspect of evaluation, grading, or feedback, the committee strongly recommends that they be transparent with their students about how and why AI is being used.

Sit with the scope of that for a second. Slides. Feedback. Grading. Those are not exotic edge cases. That is most of the week.

The reason the committee gives is not ethics in the abstract. It is that students have already noticed. The report states it plainly: students notice when instructors are clearly using AI for slides, feedback, or grading while imposing restrictions on student use, and they perceive it as a double standard.

There is a student line in the report that is more damaging than any survey number, quoted as a common refrain in the committee’s listening sessions. Why should I bother coming to class or doing the work if the teacher is just going to give an AI-generated lecture?

And on grading specifically, the report records that many students have a strongly negative reaction if, after investing their own time in an assignment, the only feedback they receive is from a machine.

This is the light bulb, and it is uncomfortable. The audit you were worried about running on your students has already been run on you, informally, all term, by people who sit in the room and read your comments. The disclosure line does not create that judgment. It just stops pretending the judgment is not happening.

The number underneath all of it

The report cites its own community’s data. In the fall 2025 Tech Survey, more than two thirds of students who responded felt AI would be important in their careers, yet only 25 percent felt MIT was adequately preparing them to use AI.

Hold those two figures next to each other, because the gap is the actual product requirement. Students are not asking to be permitted. They are saying the preparation is missing. A policy written purely as a restriction answers the first question and ignores the second one entirely.

The committee’s framing points the same way. It writes that by modeling responsible use of AI, instructors can help reinforce the new social contract the community will need in an AI infused world. Modeling is a stronger word than permitting. Modeling means the students can see how you decided.

What this costs you, honestly

Two sentences per course, plus the thinking behind them.

The student line names when AI is prohibited, allowed, or required for a specific kind of work in your subject, and says why in your discipline’s terms. The instructor line names where you use AI in producing the course, and where you do not, particularly in feedback and grading.

The second line is harder to write than the first, and not for the reason people assume. It is harder because writing it forces you to decide what you actually believe about your own practice before a student asks you in week six. Most faculty have never had to say it out loud.

The other thing worth naming: the instructor line is a credibility instrument. A policy that restricts students and says nothing about the person enforcing it reads exactly the way the MIT students said it reads. A policy that names both sides is the only version a skeptical room believes.

Write your syllabus AI policy before the guidance arrives

MIT’s instructors got a message pointing to institutional guidance. Yours may be weeks out, or may never come in a form that fits your subject. That is fine. The policy menu is not the hard part.

Take one assignment you are actually teaching this term. Write the student line for it. Write your line. Read both aloud. If you would not say the second one to a student standing in your office, that is the sentence to fix, not the one to delete.

Do that once by hand. It takes a few minutes and it will be better than anything generated for you, because you are the only person who knows what the assignment is for. Then look at your other sections and your other preps and notice that this is the same decision, repeated, in language that has to stay consistent across all of them or the students will read the inconsistency as a policy of its own. That is the moment a Syllabus Generator stops being a shortcut and starts being the thing that keeps twelve documents saying one thing. It is student-centered and accessibility-first, and it works from your Course Record, which is where your outcomes, your policies and your voice already live.

Do the first one by hand anyway.

MIT spent five months and produced a report that says, in effect, stop trying to catch them and start showing them how you decided. The syllabus was always the place that happens. It is just no longer a document about students.