Higher-Ranked Peers Push AI Rules Into Assignments

Workshop brief

Universities ranked above Iowa State have not agreed on one AI rule. They have agreed on a process. The institution supplies vocabulary and data rules, the instructor writes a course statement, and the real decisions happen assignment by assignment. In 2025–26 science and quantitative syllabi at MIT, Wharton, UW, UF and UW–Madison, the common pattern is: AI allowed for practice and learning, banned on proctored quizzes and exams, use disclosed in a line or two, and the student owns every error. Wherever an institution has set a default for a syllabus that says nothing, that default is "not allowed" (Maryland, Rutgers, Michigan State). The best student evidence shows that silence otherwise reads as permission. AI detectors have been switched off or disavowed almost everywhere.

The frontier is grounded course tutors. The two best-documented ones, Harvard's CS50 Duck and Georgia Tech's Jill Watson, both found that prompts alone cannot hold the guardrail, which is exactly what an instructor-run audit of a course tutor checks. No peer agriculture college publishes its own AI-in-teaching policy. No peer publishes an instructor workflow for "answer only from my materials." Instructors who ground a tutor in their own materials and audit it are ahead of the peers, not catching up.

Seven findings for instructors

Four quotable peer lines

On refusing questions outside the materials, Penn State Extension's grounded farm assistant Tilva (January 2026) gives the cleanest line, from David Hughes: "Where we feel confident, the AI can answer. Where we don't, we stop it." (Penn State News)

Three syllabus sentences, assembled from peer language

May be used for. You may use ISU-licensed AI tools (logged-in Copilot, Gemini or NotebookLM) or this course's study guide to review concepts, quiz yourself and get feedback on drafts, after you have first attempted the work yourself.

May not be used for, and why. You may not use AI on quizzes and exams, or to write the interpretation you submit in lab reports, because those assignments are designed to assess your own understanding and independent thinking.

Not authoritative. AI output is not an authoritative source. It can fabricate credible-looking data, references, quotations and calculations, so check anything you use against course materials. You are responsible for every error you submit.

Where each sentence comes from: - Sentence 1 draws on UW CSE 373's allow-list, Missouri BIOSC 3750's list of roles (editor, translator, idea generator, data visualization tool, tutor) and Duke STA 199's "first attempt to solve the problem yourself." - Sentence 2 draws on MIT 14.05, UF ANS 3006, and UW Econ 200's rationale (UW Econ 200). - Sentence 3 draws on Missouri microbiology and Ethan Mollick, both undated entries in Eaton's repository (Missouri's rank relative to ISU was not verified), plus UChicago Harris.

Optional fourth line: "If an assignment does not say, assume AI is not allowed. Where it is allowed, add one line: tool, what for, and how you checked it."

Four common starting points, and one upgrade for each

What Iowa State already has. ISU does publish AI syllabus guidance, with three approaches, not four. CELT's "Gen AI Syllabus Considerations," linked from its AI in Teaching page, offers "Permitted and encouraged use," "Limited use" and "Prohibited use." It adds six considerations (approach, tools, conditions, required processes, rationale, alternatives), sample language for each approach, and a step most peers skip: alternatives for students who choose not to use AI (CELT document). The required Fall 2026 ISU template asks for "Instructor policies for attendance, late work, extra credit and use of AI" (Inside Iowa State). What ISU still lacks: a default when a syllabus is silent, assignment-level labels, and a written stance on AI detectors. Start from the CELT document; the three sentences above fit inside its "Limited use" approach.

How far the evidence goes: ranks, dates and quote reliability

ISU is #116 nationally and #55 among publics in the 2027 U.S. News edition, released September 22, 2026 (Iowa Capital Dispatch via News From The States). The research team did not verify 2027 ranks school by school.

Nearly every institution discussed here sits, by wide understanding, well above #116: Stanford, MIT, Harvard, Princeton, Yale, Penn, Duke, CMU, Columbia, Cornell, Northwestern, Johns Hopkins, UChicago, Berkeley, Michigan, Washington, Illinois, Wisconsin, UT Austin, Georgia Tech, Florida, Maryland, Purdue, Ohio State, Rutgers, Minnesota, Penn State, Texas A&M, Michigan State, UC San Diego, BU, NYU, Vanderbilt and UConn. Treat that as background, not as verified fact.

Rank relative to ISU was not verified for Nebraska–Lincoln, Kansas State, Arizona State, Missouri (which the notes flag as a "near-peer") and South Carolina. They appear as land-grant or design comparators. Other sources serve only as evidence exemplars and are not ranked against ISU: - Middlebury, which U.S. News ranks among liberal arts colleges, not national universities - the SUNY system - community colleges - Chile's Universidad Adolfo Ibáñez - Turkish high schools

Several core policy texts are 2023 vintage and are labeled by date throughout: - Stanford's default for a silent syllabus (February 2023) - Harvard FAS sample statements (August 2023) - Northwestern's statements (retrieved August 2023) - CMU's contracts memo (Summer 2023) - Vanderbilt's detector shutdown (August 2023) - Kansas State's call for sample statements (August 2023)

On quote reliability (after the October 25, 2026 live re-check): - Every quote with a link was compared word for word against the raw text of the live page: HTML pages and PDFs fetched directly, and pages that block scripts (ISU, Columbia, Princeton, Harvard, Michigan, UMD, UF) read in a real browser. - Quotes that did not match were corrected to the page's exact wording or removed; the list is at the end of the report. - Lance Eaton's crowd-sourced repository was re-exported live (222 entries). It records no dates, so its entries may date from 2023–24.

Peers hand instructors a vocabulary, not a rule

Every peer leaves the permission decision to instructors. They differ in whether a syllabus statement is mandatory and in how much support the institution publishes. A few require a statement: - UT Austin: "Every syllabus must include a course AI policy." Instructors declare AI permitted, prohibited or partially permitted, and must give written disclosure instructions whenever AI is allowed (UT Austin Provost). - Nebraska–Lincoln (rank unverified): the Faculty Senate syllabus policy of November 4, 2025 requires a statement to "Describe whether AI is allowed for use in your course" (UNL Faculty Senate). - Duke: its 2026/2027 syllabus template now carries AI policy guidance from its Office of Undergraduate Education (Duke CTL).

ISU joined this small group in Fall 2026. Every undergraduate course must now carry an ISU Canvas template that includes "Instructor policies for attendance, late work, extra credit and use of AI." The provost calls completion "a component with the teaching portion of the faculty Position Responsibility Statement," and says non-compliance "will be reflected in annual reviews" (Inside Iowa State; ISU Provost).

Everyone else "strongly encourages" a statement. That includes Berkeley, Michigan, UW–Madison, Minnesota, Penn State, Maryland, Purdue, Harvard, Northwestern and Cornell's College of Agriculture and Life Sciences (CALS). Cornell CALS requires no AI statement and points faculty to Eaton's repository, a reflection worksheet and an "AI Course Policy Generator" (Cornell CALS FAQ).

ISU's gap is narrower than it first looked. The provost's template page has no AI wording, but CELT's "Gen AI Syllabus Considerations" document (found on the live re-check) supplies a three-approach vocabulary — "Permitted and encouraged use," "Limited use," "Prohibited use" — plus sample language, a list of required processes (verify accuracy, cite, reflect, "Disclose use of AI tools, prompts, generated outputs"), and alternatives for students who opt out (CELT). That puts ISU level with Berkeley and UF on vocabulary. The remaining gaps are a default for silence, assignment-level labels and a detector stance. CELT also runs a self-paced Canvas course on generative AI that "takes about two hours to complete" (September 12, 2024). In the article about it, Christine Denison's students "provide a disclosure note with every assignment to explain which AI they used, how they used it and affirm they are responsible for the result" (Inside Iowa State, 2024).

The national picture shows why support matters. Only 32% of institutions have a campus-wide AI policy, and only 23% of faculty consider such policies effective (Tyton Partners 2026). Globally, 80% of faculty report a lack of clarity about AI in teaching (Digital Education Council, Jan 2025).

The pattern is consistent. Peers have converged on a three-level course vocabulary, and the work that matters happens one level down. Berkeley's "Full AI" sample shows where grounded tools fit: "This is a GenAI-required course," with instructor-provided tools (Berkeley Senate). Ohio State's Committee on Academic Misconduct and Cornell both push assignment-level icons (OSU EHE; Cornell icons). Icons are the lightest way to fill a course-level template field without making it vague.

Silence is a policy, and peers that define it choose "no"

When a syllabus is silent, peers fall into four camps. Every institution that sets a crisp default sets it at "not allowed": - Maryland (approved January 15, 2025): "Students should assume that the use of GenAI tools to complete course assignments and assessments is not allowed unless otherwise specified in the course syllabus or assignment/assessment instructions" (UMD). - Rutgers (page modified September 1, 2026): AI tools "should not be considered permissible for coursework unless clearly stated or communicated by instructors" (Rutgers OIT).

Stanford is the outlier. Its 2023 Office of Community Standards guidance treats AI "analogously to assistance from another person" and bars using it "to substantially complete an assignment or exam" (Stanford OCS). Duke, CMU and Northwestern let their honor codes do the work: unauthorized assistance is misconduct, so silence leans restrictive. CMU's teaching center warns that unstated policies leave students unable "to effectively interpret what kinds of AI use are authorized" (CMU Eberly). No ISU default was found.

The behavioral evidence points the same way, with a caveat: it measures intentions, not behavior. Middlebury surveyed students from December 2024 to February 2025 (analysis n=634). 42.2% said they would likely use AI under no explicit policy, nearly the same as the 40.9% under "allowed if cited." Under a ban the figure fell to 13.4%, about 39 points below unrestricted use (Contractor & Reyes, arXiv).

EDUCAUSE's 2025 student survey found: - Students whose instructors prohibit AI are significantly more likely to say they do not use it. - Only 51% of students had received any AI guidance. - 52% said most or all of their instructors prohibit AI.

Source: EDUCAUSE 2025 via ACCJC.

A default alone is not enough. Middlebury students asked for specific examples of acceptable and unacceptable uses. Interview research found students using more than 20 rationalizations, and found their practice drifting from the instructor's intent at "at least five distinct sites" (Kim et al., arXiv 2026).

The design that answers both problems pairs a "no unless stated" default with an explicit allow-list. UW–Madison Biology 151 does exactly this, and its policy opens on an encouraging note: "We are excited about the use of AI within the fields of education and biology" (Bio 151 syllabus).

2025–26 syllabi split rules by assignment and shrink disclosure

The evidence comes from four sources: - Thirteen 2025–26 syllabi fetched directly from biology, animal science, statistics, economics, chemistry and intro CS courses. - Lance Eaton's crowd-sourced repository: 222 entries as of October 25, 2026. Writing courses dominate (39 entries). STEM, health and agriculture together are only about 15%, and only one entry is agriculture (Eaton sheet). - A peer-reviewed coding of 92 instructor policies: 52 allowed restricted use, 27 encouraged use with restrictions, 12 were zero-tolerance and 2 were unrestricted. About 90% placed responsibility for AI competency on the student (Tong et al. 2025 via NCA). - A SIGCSE 2025 study of 98 computing syllabi at 54 R1 universities. Most prohibited or restricted AI, and most placed the guidance inside the academic-integrity section (arXiv 2410.22281).

The dominant 2025–26 STEM pattern protects the proctored assessment rather than banning the tool: - MIT 14.05 "explicitly encourages" AI, requires students to cite it, and bans it on in-class exams (MIT OCW). - Wharton BEPP 6120 encourages experimentation, with a "closed book" and "closed computer" final as the backstop (Wharton). - UF ANS 3006 Online, the one animal-science example with an AI statement, says: "AI should not be used to assist with any quizzes. You can use AI for preliminary brainstorming on written assignments." It asks students to name the platform (UF ANS 3006). - UW CSE 373 lists allowed uses, for example "Generate code to complete tasks that you have already mastered from previous coursework," alongside banned ones (UW CSE 373). - Duke STA 199 separates code from prose. Students must "first attempt to solve the problem yourself" and may not paste AI-written narrative verbatim. Its principle: "Working with AI should not reduce your thinking ability" (STA 199). - Harvard CS50 permits only its own guarded tools: "Using CS50's own AI-based software, including the CS50 Duck (ddb)" is reasonable, while third-party AI is not (CS50 syllabus).

A restrictive minority persists, usually as department boilerplate. UF ECO 3203 bars AI on all graded work "unless their instructor has expressly granted that permission" (UF ECO 3203). UW Econ 200 is short and gives its reason: the assignments "are designed to assess your understanding and independent thinking" (UW Econ 200).

Silence and passing mentions persist too, sometimes inside one course. Georgia Tech CHEM 1211K's Fall 2026 lecture syllabus never mentions AI, while its lab syllabus says "Assignment submissions containing AI-generated text will be considered a violation of academic integrity" and that AI used to revise writing "should be disclosed and cited for transparency" (lecture; lab). An earlier section page that researchers quoted now returns "permission is required," so that quote was dropped. Silence is common even in strong agriculture departments. Only one of six publicly posted 2025 UF Animal Sciences syllabi has an AI statement, and the in-person section of ANS 3006, taught by the same instructor in the same term, has none (UF ANS 3006 in-person). No 2025–26 top-100 syllabus from food science, agronomy or agricultural economics with fetchable AI text turned up, so CALS-discipline examples remain a real gap.

Survey data put restrictive policies in the 2024 mainstream, with the trend moving away from them. In Ithaka S+R's February–March 2024 survey, 42% of US instructors completely prohibited student AI use (an earlier draft added "45% in the sciences"; the report's 45% figure is humanities faculty skepticism, so it was removed) (Ithaka S+R, June 2024). Tyton's June 2026 report found 47% of faculty have redesigned assessments, and 24% are "Integrators" who use oral exams, drafts and student critiques of AI output. Faculty rate AI-embracing policies effective twice as often as bans (Tyton 2026). That is a perception, not an outcome measure, but it is persuasive with skeptics.

Disclosure now runs both ways: - Cornell: "When using material generated from an LLM in course materials or in assignments that students submit, transparency is key, and these instances should be properly referenced" (Cornell CTI). - Minnesota offers an optional instructor-disclosure statement (Statement C) (UMN). - Yale and Johns Hopkins encourage instructors to tell students when AI helped build course materials (JHU CTEI). - UW Bothell's intro AI course goes furthest. Its student-ratified "Student Bill of AI Rights" requires the instructor to personally complete any AI-generated assignment before issuing it (Pisan, arXiv 2608.05175).

That last rule is the five-test audit restated as an ethics commitment, and it belongs on any instructor-built study tool.

Detectors are gone; data classification does the governing

Peers are nearly unanimous on detectors. ISU has no written stance on AI detectors; CELT's guidance mentions only Turnitin Draft Coach as a student plagiarism self-check. Each workshop syllabus should simply leave detectors out.

Penn's guidance adds the most citable line for skeptics. Instructors "should not upload the student's work into an 'AI detector'," and "Uploading a student's work without their consent is a violation of FERPA and Penn's privacy policies" (Penn CETLI). In place of detection, peers point to assessment design: Berkeley recommends in-person exams and oral explanations (Berkeley Senate).

Data governance works by approving specific tools for specific data-classification levels. There are four models.

ISU's own rules are the most permissive in this comparison, if read literally. ISU says that when logged in, "you can use it for confidential data" for Copilot, and Gemini and NotebookLM can be used "for confidential data too," while free ChatGPT is limited to data "classified as low" and "The university is working to add a license" (ai.iastate.edu). ISU's data page defines its four classes as Restricted, "High. Confidential student information protected by Family Educational Rights and Privacy Act (FERPA)," Moderate and Low (ISU Data Services). Read together, "confidential" points at the FERPA class, which Harvard, Illinois and Penn State all exclude from the same tools. The AI page never says so explicitly, so confirm with ITS before any student work goes into a tool, and keep instructor-built tools to instructor materials.

Peers differ on that same tool. UIUC clears Gemini and NotebookLM only for "Internal and Public" data (Illinois). Penn State clears its Gemini Notebook only for Levels 1–2 (PSU).

Stanford's tool matrix shows how fast these approvals move (checked live October 25, 2026). Claude Code and Claude Cowork are approved when provisioned through "the UIT managed workspace, which was released on June 30, 2026," but the Claude for Education entry reads: "Although Anthropic has redeployed Claude Fable 5, this tool remains disabled in the Stanford enterprise environment as it does not currently align with the university's obligations for data governance, research protection, and institutional risk management" (Stanford UIT). The lesson for ISU: approval is per tool and per data class, and it can change mid-semester.

Across peers the risk boundary is clear: an instructor's own syllabus and lectures are rarely the problem; student work is. Harvard FAS (2023): "Faculty must get documented permission from students before putting original student content into any generative AI tool" (Harvard OUE; the page moved from its old address). The same page now announces that "Starting in Fall 2025," faculty can use the Respondus lockdown browser for in-person exams "to ensure that students do not use AI unless the course asks them to do so." CMU says FERPA protection begins once an instructor accepts an assignment for grading (CMU Eberly).

Peers have not settled the question of instructor materials. Purdue's teaching guidance says instructor-generated materials "should never be uploaded to any third-party site" (Purdue OVPTL), yet its tool list clears iSeek.ai for "course curricula and materials" (Purdue). Yale protects faculty IP from the student side, with a sample clause: "Do not submit instructor-created materials to AI tools without explicit permission" (Yale Poorvu). No US policy found says who owns or may reuse course materials once faculty upload them to an institutional AI tool.

Grounded tutors work only when someone audits them

Course-grounded tutors at peer universities fall into three tiers. The middle tier, a campus build-your-own platform with a Canvas connector, is the dominant fall 2026 model and the closest analogue to what an instructor can build today.

The guardrail lessons are blunt. Malan's team set out to build a "less-helpful version of ChatGPT," found prompting insufficient, and added code that screens every response. Even so, the Duck emitted code blocks "up to a quarter of the time" before fixes, and some students sent about 200 questions each (Harvard Gazette, Sept 2026). Kestin's physics team found that "a system prompt could not reliably provide enough structure to scaffold problems with multiple parts," so the platform enforces it (Kestin et al. 2025). The cheapest guardrail an instructor controls is what goes into the index: UCSD lets instructors leave assignments and study guides out, and Maizey defaults to content students can already see.

The outcome evidence is thin but consistent: - Harvard physics tutor (crossover RCT, N=194): 0.63 SD over in-class active learning, with 0.73–1.3 SD by quantile regression. The tutor was heavily structured, built on expert solutions, and had authors who overlapped with the instructors (Kestin). - Chilean econometrics course (two pre-registered RCTs, about 300 students): a course-grounded GPT had no effect when merely offered (0.05 SD). Coaching students to use it tutor-style produced +0.22 SD and raised the pass rate 12.8 points. This is a preliminary working paper (Gallegos, IZA 2026). - Turkish high schools: unguarded GPT-4 left students 17% lower on unaided exams, while a hint-first tutor roughly neutralized the harm (Bastani et al., PNAS 2025). - Biology-textbook study (online RCT, N=124): the text-only grounded chat was rated most engaging but scored lowest on the post-test (Taneja et al. 2026).

So satisfaction surveys say nothing about learning. No RCT of an instructor-built Canvas-connector bot exists.

Each of the five audit tests maps onto a documented peer failure.

The agriculture-specific precedent is Extension, not teaching. Penn State Extension's Tilva launched in January 2026 after about six months of testing with 29 program teams. More than 260 educators in 67 counties oversee it. It draws preferentially on vetted materials and sends pesticide, food-preservation and animal-medication questions to humans (Penn State News). Other Extension services publish AI rules: Texas A&M AgriLife Extension's formal procedure (February 2025) (AgriLife), UW–Madison Extension's guiding principles (June 2025) (UW Extension) and Illinois Extension's communications guidance (January 2025; its text could not be extracted on the re-check, so the earlier "Always disclose" quote was dropped) (Illinois Extension). Their shared vocabulary of human oversight, research-based content and disclosure transfers directly to course tutors.

UT Austin supplies a ready "where AI does not belong" example. Under Texas law, AI must not "autonomously make or be a controlling factor in making consequential decisions without meaningful human review," and the university lists grading among those decisions (UT Austin).

Three gaps matter for the workshop: - No peer publishes an instructor workflow for grounding, checking and scoping a tutor. Researchers found none at any of the private universities they covered or at any land-grant ag college. - No peer gives guidance on testing the accuracy of an instructor-built tutor. - No peer page states how long prompts, indexes or uploaded files are retained.

Teaching AI: elite intro courses lag on LLM applications

Top technical intro-AI courses still teach the canon from Russell and Norvig and Berkeley's Pac-Man projects. They require data structures plus probability and put most of the grade on exams. AI 2010 belongs with a different peer set: any major, no CS prerequisite, sections of about 100, project-based.

AI-use policies inside AI courses show the clearest recent shift at any top program. Berkeley CS 188 moved from "Absolutely Forbidden" in Fall 2025 to "GenAI is a consulting tool, not an author" in Fall 2026, with a reflection on AI use in every assignment (Fa25; Fa26; both re-checked on the live pages October 25, 2026).

LLM content in flagship courses remains incremental: one lecture plus maybe a homework (Berkeley's transformers lecture, Stanford's "Language Models" and "AI Supply Chains"). No verified top-ranked intro AI course has a RAG or LLM-application project. RAG appears only in graduate courses, such as Stanford CS 224V, where students reduce hallucination with RAG in a project worth 65% (CS 224V). It also appears in Purdue's non-credit AI Forge, whose projects include a course-materials Q&A bot whose answers TAs verify (Purdue AI Forge).

Ethics is most formalized at Stanford. CS 221 has a graded "AI & Society" homework, two lectures on societal issues and four ethics course assistants (CS221). The published course-design literature is mostly single-course experience reports. Pisan's redesign, CS50's SIGCSE 2024 paper and a 2026 interactive intro-AI report (Reddig et al.) lead it. EAAI-27 meets February 21–23, 2027 (EAAI-27). A grounded-and-audited AI 2010 project would be ahead of every verified peer.

AI for all students: Purdue requires, Ohio State embeds, Florida credentials

Two Big Ten land-grants now require AI of every undergraduate, using opposite designs: - Purdue's "AI working competency" was approved in December 2025 and applies from Fall 2026 entrants, transfers included. Students take one competency course of 1–4 credits chosen by major. Purdue reports 22 courses developed or revised and 300+ plans of study updated (Purdue, Aug 2026). The 74 reported 35 courses "identified," a discrepancy that remains unresolved (The 74). The course list mixes CS 173 AI Competency, SCLA 19560 AI Literacy, FIN 258 and BTNY 110 Introduction to Plant Science (Exponent). - Ohio State's AI Fluency puts generative-AI basics into the required General Education Launch Seminar and First Year Success workshops. Every college has a roadmap that builds AI into existing degrees "without adding credits," reaching about 53,000 undergraduates, starting with the Class of 2029 (OSU OAA, Jan 2026).

On minors, ISU compares well. Nationally, the New York Times count cited in July 2026 was 74 AI majors and 89 AI minors at US colleges (Altoona Mirror).

UMD's AGNR 230, "Artificial Intelligence: Transforming Food, Agricultural, and Environmental Systems," shows what a gen-ed agriculture AI course looks like: 3 credits, 50 seats, about 45 enrolled for Fall 2026, co-taught (UMD Testudo). The agricultural AI Institutes' education arms consist of research experiences (REUs), bootcamps such as AIFS AIBridge, K-12 programs and modules (AIBridge; AIFARMS). None operates an undergraduate AI-for-agriculture minor. ISU hosts AIIRA, the AI Institute for Resilient Agriculture, supported by NSF and USDA-NIFA award #2021-67021-35329 (AIIRA), an asset UF and UMD lack.

Faculty-development comparators match the CALS series' intent but not its hands-on, artifact-producing format: - Ohio State's Drake Institute runs an application-based AI-Infused Course Design Institute and a "Teaching and Learning to Build AI Fluency" endorsement, and the university offers AI Fluency Innovation Microfunds (OSU Drake). - ASU's Assessment Design Accelerator targets faculty in its top-100-enrolled courses (ASU). - Duke's AY 2024–25 AI Jump Start grants paid up to $7,500 plus a $1,000 stipend (Duke CTL). - Cornell's AI Integration Pilots embed a graduate student or postdoc in faculty teams from spring 2027, funded by a $2M gift (Cornell Chronicle).

Recommendations

(1) For a faculty workshop on grounded course tools

(2) AI 2010's own policy and design

AI 2010 should model "required to use, required to understand." Peers that teach AI while banning it are handling a different problem: protecting practice at implementing algorithms. An applied course for any major is closest to Berkeley's Fall 2026 policy, CS50's sanctioned tutor and Pisan's redesign. One constraint from CMU and Yale applies: required tools must be university-licensed, never personal accounts (CMU Eberly; Yale Poorvu). A draft drawn from those peers:

AI in AI 2010. In this course you are required to use AI and required to understand it. Every required tool is ISU-licensed, so you never need a personal account. Use ISU tools and the AI 2010 course tutor on project work, including code, as a consulting tool, not an author. End each submission with a short AI note (tool, what you used it for, a link to your key prompts, how you checked the output) and one or two sentences on whether AI helped or hurt your learning. In-class checks and your project defense are AI-free, because they show what you can do and explain on your own. AI output is not authoritative: you are responsible for every claim, number and line of code you submit, and you must be able to explain it. We do not use AI detectors; we use conversations, process evidence and your defense.

(3) ISU and CALS positioning

Conclusion

The peer frontier has moved past permission and is moving toward verification. Higher-ranked universities have largely settled what to say: a three-level vocabulary, a "no unless stated" fallback, short disclosure, and no detectors. None of them yet tells faculty how to check what an AI built from their own materials actually says. The systems with real evidence behind them, the CS50 Duck, Jill Watson and the Harvard physics tutor, earned it through checks and structure layered on top of prompts. The RCT literature says the same about students: access alone does nothing, and guided use helps. That makes the five-test audit more than a workshop exercise. It is the missing institutional practice, and ISU could standardize it before any peer does.

ISU's position is unusual. It is one of very few universities that require an AI statement in every undergraduate syllabus, and CELT already supplies a three-approach vocabulary with sample language. What is missing is a default for silence, assignment-level labels, a detector stance and a plain statement of which tools may hold FERPA data. That gap is an opening for CALS rather than an embarrassment. Pairing the Fall 2026 mandate with CELT's language, an honest data table and a verification practice would let ISU lead on the part of AI-in-teaching that peers have not reached. AI 2010 is the place to prove it with students.

What changed after the live re-check (October 25, 2026)

About 90 linked quotes and 180 figures were compared against the raw text of the live pages. Most matched exactly, including all four quotes and the Tilva line in the workshop brief. These are the items that changed:

Still not re-checkable from outside ISU: the ISU Canvas syllabus template itself (Box, ISU login), so its AI-use field wording was not re-checked.

Sources

Iowa State baseline - U.S. News 2027 ranks: Iowa Capital Dispatch via News From The States - ISU Provost: Required Syllabus Templates - Inside Iowa State: New undergraduate syllabus templates (Apr 2026) - ISU: Safely Use AI - ISU Data Services: Data classification - ISU CELT: AI in Teaching and Learning - ISU CELT: Gen AI Syllabus Considerations (Box document) - Inside Iowa State: CELT course assists faculty with AI (2024) - ISU Catalog: Applied AI minor

Institutional frameworks, defaults and disclosure - UT Austin Provost: AI policy - UNL Faculty Senate syllabus policy (Nov 2025) - Duke CTL: AI policies in syllabi - UC Berkeley Academic Senate GenAI guidance (2025) - UF Provost: Guidelines for Use of AI at UF (Version 1, 2023) - UW–Madison CTLM AI statements - UW CTL sample AI statements - UMN Provost GenAI syllabus statements - Penn State eLDIG statements - Cornell CTI: AI & academic integrity - Cornell AI course policy icons - Cornell CALS FAQ for educators - Michigan State CTLI syllabus guide - Maryland Academic Guidelines (Jan 2025) - Rutgers OIT AI guidance - Ohio State EHE policy guidance - Purdue OVPTL AI guidance - Stanford OCS generative AI guidance (2023) - Stanford Teaching Commons: course AI policy - Harvard OUE Generative AI Guidance (2023 FAQ, moved; Fall 2025 Respondus note) - Princeton Scholarly Integrity: disclosing AI - Northwestern AI in courses - CMU Eberly AI tools - Penn CETLI generative AI - Columbia Provost AI policy - JHU CTEI teaching and AI - Yale Poorvu: protecting IP and data

Course syllabi and policy banks - Georgia Tech BIOL 1207L (Fall 2026) - UW–Madison Biology 151 (Fall 2025) - UF ANS 3006 Online (Fall 2025) - UF ANS 3006 in-person (Fall 2025) - UF ANS 3043 (Fall 2025) - Duke STA 199 (Fall 2025) - Harvard CS50 syllabus (Spring 2025) - UW CSE 373 GenAI policy (Winter 2025) - MIT 14.05 (Fall 2025) - UChicago Harris PPHA 31202 (Fall 2025) - Wharton BEPP 6120 (Fall 2025) - UF ECO 3203 (Fall 2025) - UW Econ 200 (Summer 2025) - Georgia Tech CHEM 1211K lecture syllabus (Fall 2026) - Georgia Tech CHEM 1211 lab syllabus (Fall 2026) - Eaton, Syllabi Policies for Generative AI repository

Surveys and behavioral evidence - Contractor & Reyes, Middlebury (arXiv 2508.00717) - EDUCAUSE 2025 Students and Technology Report (ACCJC copy) - Ithaka S+R instructor survey (June 2024) - Tyton Partners Time for Class 2026 - Digital Education Council faculty survey (Jan 2025) - Tong et al. 2025 via NCA - Computing syllabi at R1s (arXiv 2410.22281) - Kim et al., "It's OK Because…" (arXiv 2605.29090)

Detectors and data rules - MSU: Turnitin AI detection - Yale Poorvu: Turnitin - Vanderbilt: disabling Turnitin AI detector (2023) - AI Weekly secondary report (unverified) - pasqualepillitteri.it secondary report (unverified) - Purdue Authorized AI Tools (Oct 2026) - Ohio State tools by data level - Berkeley licensed AI tools - Illinois GenAI apps - Harvard HUIT AI tools - MIT IS&T AI tools - Stanford UIT AI matrix - Penn ISC AI tools - TAMU AI Chat - Duke Today: Duke-managed AI services - Penn State Gemini Notebook

Course-grounded tutors and evidence - Liu et al., Teaching CS50 with AI (SIGCSE 2024) - Harvard Gazette: Taming the Duck (Sept 2026) - Georgia Tech: Jill Watson outperforms ChatGPT - Kestin et al., Scientific Reports 2025 - Bastani et al., PNAS 2025 - Gallegos, IZA DP 18513 (2026) - Taneja, Singh & Goel (arXiv 2604.02221) - U-M Maizey Canvas connector KB - U-M Provost: AI services in the classroom - U-M GPT system prompts - UCSD TritonAI instructional AI - UIUC Illinois Chat Canvas KB - Illinois Computes: Illinois Chat - Duke CTL: MyGPT Builder - ASU CreateAI (Mar 2026) - UF NaviGator launch (2024) - Purdue RCAC GenAI Studio - OpenAI study mode (2025) - Anthropic: Claude for Education (2025) - Google: Gemini for higher education (2025) - Penn State Extension Tilva (Jan 2026) - Texas A&M AgriLife Extension AI procedure (2025) - UW–Madison Extension AI principles (2025) - Illinois Extension AI guidelines (2025)

Intro AI courses - Stanford CS 221 (Autumn 2025) - Berkeley CS 188 policies, Fall 2025 - Berkeley CS 188 policies, Fall 2026 - CMU 15-281 - CMU 07-180 - CS50 AI honesty - UIUC CS 440 (Fall 2025) - GT CS 3600, Europe section (Spring 2026) - Penn State "Artificial Intelligence: The Very Idea" - Ohio State ENGR 2194 - MIT 6.S191 - Stanford CS 224V - Purdue AI Forge - Pisan, UW Bothell CSS 382 redesign (arXiv 2608.05175) - Reddig et al. (arXiv 2603.28679) - EAAI-27 CFP

AI literacy, requirements, minors and faculty development - Ohio State AI Fluency launch (2025) - Ohio State college roadmaps (Jan 2026) - Purdue AI competency requirement (Aug 2026) - The 74 on AI graduation requirements (Feb 2026) - Purdue Exponent on AI competency credit (Aug 2026) - ASU Provost Q&A (Aug 2026) - UF AI certificate catalog - UF Engineering: Charting a path forward with AI (2025) - UF CALS Artificial Intelligence - UMD AGNR 230 (Fall 2026) - UMN AI literacy courses and minor (Mar 2026) - GeekWire: UW AI minor (June 2026) - Michigan Engineering: AI minor (Nov 2025) - Georgia Tech AI & ML minor (2024) - Cornell Digital Agriculture minor - Purdue Data Driven Agriculture minor - Cornell AI for All (2026) - Cornell Chronicle: Everyday AI course - Cornell Chronicle: AI Integration Pilots (Sept 2026) - Penn State AI Essentials - Inside Higher Ed: SUNY adds AI to gen ed (2025) - Altoona Mirror: AI majors and minors count (July 2026) - AIFS AIBridge Boot Camp - AIFARMS education and outreach - AIIRA: AI Institute for Resilient Agriculture - Ohio State Drake Institute AI fluency support - Duke AI Jump Start Grants (AY 2024–25)