Four in Five Say No

We classified the AI policy in 2,719 college syllabi. The result is consistent across every recent segment of the DormWay corpus: most professors still make “no” the default.

There is more agreement than the public discussion suggests.

The conversation about AI in higher education often focuses on AI literacy, responsible use, and the limits of detection. Course-level policy is much less permissive. In the latest DormWay analysis, almost four in five syllabi that address AI either prohibit it or make prior authorization the condition for using it.

DormWay analyzed 4,470 processed syllabus documents uploaded by 1,667 users between March 1 and August 20, 2026. The parse output contains 828 distinct institution names after basic case and whitespace normalization. We examined policy text extracted from the syllabi themselves and excluded generated advice fields.

Of those 4,470 documents, 2,719, or 60.8%, contain an explicit AI reference in the extracted policy text.

The important result appears inside the 2,719 syllabi that mention AI.

The default is still no

The restrictive group contains 2,158 syllabi. Some impose a blanket ban:

“No use of AI is permitted. The use of Artificial Intelligence tools is strictly prohibited in this course.”

— THR 3550, Saint Louis University

“The use of AI tools is prohibited for assignments in this course.”

— HR-3650-050, University of Winnipeg, Global College

Other policies use a permission gate. AI may be available in theory, but a student cannot proceed until the instructor authorizes it for a particular assignment. That is still a restrictive rule because the default remains no.

This distinction matters. Disclosure language and prohibition language often appear in the same paragraph. A policy may require a citation after the instructor grants permission. The permission requirement is the binding condition.

Disclosure is a minority position

The disclosure group contains 351 syllabi, or 12.9% of those that mention AI. These policies let students use AI without seeking permission first, provided they document the use.

“The use of Generative Artificial Intelligence (GAI) is permitted on this course with full disclosure.”

— ME1212, Louisiana State University

“AI usage must be disclosed or will be treated as plagiarized work.”

— CWE 101, North Seattle College

These policies are easier to follow because the student can determine the rule from the syllabus. They remain uncommon. Restrictive policies outnumber disclosure policies by more than six to one.

Open policies remain rare

Only 38 of the 2,719 AI-mentioning syllabi, or 1.4%, are open or encouraging without a prior-permission or disclosure requirement.

“Students are encouraged to integrate AI responsibly to enhance their work rather than replace original thought and effort.”

— 401, Northwestern University

The open group is larger than it was in the July snapshot, but its share is unchanged at 1.4%. The latest data still does not support the idea that open-ended AI use has become a normal course policy.

Ambiguity has not disappeared

Another 172 syllabi, or 6.3%, mention AI without providing an actionable course-level rule. They commonly refer students to another policy or learning-management page.

“See Folio for specific guidelines and working definitions.”

— BIOL 2252L–AK, Georgia Southern University

“The course adheres to the Mandell School’s policy on the Use of Artificial Intelligence (AI).”

— SASS 528, Case Western Reserve University

The reference may point to a valid rule, but the student still has to find it somewhere else.

The same-campus problem remains

The restrictive majority does not eliminate course-level inconsistency. Among the 276 named institutions with at least three AI-mentioning syllabi in the current corpus, 141, or 51%, contain more than one stance.

The current 51% figure should not be read as a change in institutional policy. The corpus now contains more syllabi per institution, which makes course-level differences easier to observe.

What this means

Faculty policy is substantially more restrictive than the broader discussion about AI in education. In the syllabus, the prevailing instruction is simple: students should assume AI use is prohibited unless the course states otherwise.

The remaining problem is retrieval. A student taking five classes can still face five different implementations of that rule, with some policies buried in a separate page or assignment description. That is why DormWay treats policy extraction as part of syllabus processing. The useful job is to show the deadline, grading weight, and applicable rule before a student makes a decision.

Methodology

Corpus. The analysis covers 4,470 syllabus documents processed between March 1 and August 20, 2026. It starts with braingains_documents, filters to document_type='syllabus' and successful states (completed or indexed), joins each document to service_data through the shared workflow ID, and retains the latest parse for each document. The unit is a processed document, so separate uploads or revisions can appear separately.

Why the March start date. Earlier extractor versions returned much less policy text. Mixing those records into the current window would undercount older AI references and create a false trend. Average extracted policy volume is reasonably stable from March onward. This is a cumulative observational corpus, so the 60.8% mention rate should not be interpreted as a time trend or a representative estimate for all colleges.

AI detection. Detection runs only against extracted policy_text and summary fields, plus the academic-integrity, technology, and other fields in the older object format. Generated consequences, recommendations, and action items are excluded. Search terms include ChatGPT, generative AI, artificial intelligence, AI, Copilot, large language model, and LLM. One Adobe Illustrator file-extension false positive was explicitly excluded.

Classification. Each of the 2,719 AI-mentioning syllabi was classified according to one question: What is the default for a student who wants to use AI on course work? Blanket bans, permission gates, assignment-specific authorization, and prohibitions on submitted work are restrictive. Broad use with a citation or disclosure requirement is disclosure. Broad use without those requirements is open. References that do not state a usable rule are ambiguous.

Calibration. The July analysis used a classifier validated against 50 labels assigned by an analyst before the model run, with 90% agreement. The current structured classifier reproduced the reviewed July restrictive share within 0.6 percentage points: 79.1% versus 78.5%. It then returned 79.5% restrictive for records added after the July review and 79.4% for the combined window.

Limitations. DormWay users choose which syllabi to upload, institutions are represented unevenly, and 919 documents do not contain a university name in the parse output. Another 240 documents contain no extracted policy text and therefore fall into the no-detected-reference group. Institution counts use basic name normalization rather than a fully resolved institutional identity table. Syllabus language may also differ from what an instructor permits in practice.