University Leadership Brief

Executive Summary

The Enterprise Deal Is the Policy: What Leadership Owes Shared Governance Before the Next Signature

Our analysis of 6,252 sources this week surfaces a strategic dilemma your peers are resolving by procurement rather than policy: system-wide AI contracts are being signed before faculty senates, IRBs, or student governments have ruled on the pedagogical questions those contracts foreclose. The Cal State–OpenAI agreement is now drawing organized refusal from students and faculty who say they were not consulted Cal State struck a deal with OpenAI. Some students and …. At Arizona State, faculty are pushing back on an AI Course Builder that compresses curriculum design into a vendor workflow Faculty Concerned About ASU’s New AI Course Builder. Surrey, meanwhile, is embedding AI into every degree from September 2026 — a discipline-by-discipline mandate that puts the integration question downstream of the strategic one Surrey embeds AI in every degree from 2026.

The strategic challenge is not whether to adopt. It is that the EULA, the API contract, and the SSO integration are now doing the work that academic policy used to do — defining acceptable use, data flows, and pedagogical scope — and they are doing it faster than your governance cycle can review. The Adelphi AI-plagiarism lawsuit shows the downstream cost when detection policy outruns due process An Adelphi University student was accused of using AI to … - Newsday. South Africa’s national AI policy citing fabricated, AI-generated sources shows the cost when speed outruns verification at the policy layer itself South Africa’s AI policy cited fake research, created by AI.

This briefing provides the governance framework options with implementation evidence AI Leadership in Education: A Governance Framework to Scale Safely, the documented failure patterns to avoid before your next vendor signature, and the resource implications — legal, instructional-design, and assessment — your cabinet needs before the contract becomes the policy.

Critical Tension

Leadership Brief: The Vendor Is Already Writing Your AI Policy

The Strategic Dilemma

The institutional contradiction this week is not abstract. It is optimizing for efficiency and scalability versus preserving and fostering deep cognitive processes — and it is now showing up as signed contracts. Cal State’s system-wide OpenAI deal has produced open refusal from students and faculty who were not consulted on what is, functionally, a curricular decision dressed as an IT procurement Cal State struck a deal with OpenAI. Some students and …. At ASU, faculty are objecting to an AI course builder that compresses the design judgment of a tenure-track instructor into a generation pipeline Faculty Concerned About ASU’s New AI Course Builder. Surrey has gone further, committing to embed AI in every degree from September 2026 Surrey embeds AI in every degree from 2026. Each of these is a governance choice that crossed shared-governance lines before shared governance got to vote.

The reason this is a hard strategic problem — not a “we need more data” problem — is that the efficiency case and the cognitive-development case do not share a measurement vocabulary. FTE, time-to-degree, and retention dashboards reward scaling. The thing being traded away — the disciplined struggle that produces a writer or an analyst — does not appear on the same dashboard until the assessment cycle three years later, by which point the contract is renewed. Harvard’s faculty are explicit that the issue is preserving the difficulty of learning, not adding more tools Preserving learning in the age of AI shortcuts — Harvard Gazette. Future Shock named this asymmetry decades ago: institutions whose decision rhythms are annual cannot govern technologies whose model versions ship quarterly.

Why Peer Institutions Aren’t Helping

The sector is not converging. Surrey is mandating embedding; Cal State is bulk-licensing; Staffordshire ran a course “taught in large part by AI” and got a student revolt ‘We could have asked ChatGPT’: students fight back over course taught by AI; ASU is automating course design over faculty objection. There is no peer benchmark — there is a policy spread, and the spread itself is the evidence that nobody has solved this.

Copying a peer carries hidden liability. South Africa’s national AI policy was found to cite fabricated research generated by AI, a governance failure at the document-production stage South Africa’s AI policy cited fake research, created by AI. Adelphi is being sued by a student wrongly accused of AI use, the kind of case now accumulating into a documented lawsuit pattern An Adelphi University student was accused of using AI to … - Newsday AI Detection Lawsuits: Every Student Case, Outcome, and What the Data …. A policy borrowed from a peer who has not yet been sued is not a tested policy; it is an untested one with a logo.

What Complicates Navigation

The discourse leadership is reading is structurally lopsided. Across the 2,287 HE-tagged articles this week, student voice surfaces in only 3.76% of coverage, parent voice in 0.29%, critic voice in 0.29%, and vendor voice in 0.29%. The vendor number looks small until you notice that vendor framing does not need volume — it travels through procurement contracts, through ChatGPT Edu deployment guides ChatGPT Edu at OpenAI - OpenAI Help Center, and through the IT-side language of “deployment,” “rollout,” and “scale” that arrives at cabinet meetings already pre-framed. The students who actually use these systems, and the parents paying tuition that increasingly funds their licenses, are barely in the record being consulted.

The dominant metaphor — AI as “tool” — is what makes this invisible. A tool framing routes the decision to IT procurement and treats the pedagogical question as downstream. But the Yale CELI reading of the entry-level labor market suggests the cognitive scaffolding being automated away is precisely the scaffolding new graduates used to climb AI won’t kill your job — it will kill the path to your first one. MIT Sloan’s “persuasion bombs” framing — generative systems exerting directional pressure on user judgment — is closer to what is actually being licensed How generative AI ‘persuasion bombs’ users. A leadership team that signs an enterprise license under a “tool” frame has, without naming it, also signed a curricular policy, an academic-integrity policy, and a labor policy. The governance question is whether those three were ever on the agenda.

Actionable Recommendations

Leadership Brief: Five Moves That Don’t Survive Contact With Faculty

The dominant leadership posture this spring — sign the enterprise deal, embed AI across the curriculum, mandate the course builder, deploy the detector — is producing a remarkably consistent failure pattern: announcements made without the people who teach, followed by a slow, expensive walk-back. The week’s evidence is unusually clean on this point. Below are five recommendations grounded in what’s actually breaking.

1. Stop signing system-wide vendor deals before shared governance has touched them.

The common move is to negotiate an enterprise license — ChatGPT Edu, a Microsoft Copilot tier, a Gemini for Education seat — at the chancellor or system level, then announce it as a fait accompli. This is failing publicly. The Cal State system’s OpenAI deal is now generating organized refusal from faculty and students who say they were never consulted and object to being defaulted into a tool they don’t trust Cal State struck a deal with OpenAI. Some students and faculty refuse to use it. ASU’s AI Course Builder is drawing similar faculty objections about who decides what counts as a course Faculty Concerned About ASU’s New AI Course Builder. The hidden complexity: enterprise contracts encode pedagogical defaults — model choice, retention windows, opt-out architecture, “academic integrity” telemetry — that the senate never voted on.

Recommended alternative: route any system-wide AI procurement through the same shared-governance path you would use for an LMS migration, before signature, with binding faculty-senate review of the data, IP, and pedagogy clauses. Vendors will resist this because it slows the deal; that’s the point.

Implementation framework: - Phase 1 (Month 1–2): Inventory existing AI contracts at the unit, college, and system level. Most institutions don’t know what they’ve already signed. - Phase 2 (Month 3–4): Establish a standing AI procurement review (provost office + senate + CIO + general counsel + a student rep with vote) with authority to reject or amend before execution. - Phase 3 (end of academic year): Public registry of all AI vendor contracts, with data-handling and pedagogical-default summaries, accessible to faculty and students.

Required resources: ~0.5 FTE legal/contracts analyst; existing senate committee bandwidth. Success metrics: zero new system-level AI contracts signed without documented senate review; published vendor registry. Risk mitigation: vendors will offer “pilot” structures to bypass governance — treat pilots as procurements.

2. Replace the AI detector with assessment redesign — and budget for it like a curriculum project, not a software purchase.

Spending on Turnitin’s AI detector, GPTZero, or Copyleaks is the obvious move and is now actively producing legal liability. Adelphi University is being sued by a student who says he was wrongly accused based on detector output An Adelphi University student was accused of using AI to …, and the broader docket of detection lawsuits is growing AI Detection Lawsuits: Every Student Case, Outcome, and What the Data Shows. The arms race has produced its mirror image: students now run their own writing through “humanizers” specifically to evade detectors To avoid accusations of AI cheating, college students turn to AI. The empirical literature is converging that detection is an unreliable basis for academic-integrity adjudication Contra generative AI detection in higher education assessments.

Recommended alternative: fund authentic-assessment redesign at the department level — oral defenses, scaffolded drafts, in-class synthesis, process portfolios — and explicitly retire detection-based adjudication Beyond Detection: Redesigning Authentic Assessment in an AI …, Reimagining Writing Assessment for the AI Era.

Implementation framework: - Phase 1 (Month 1–2): Issue interim guidance that AI-detector output alone is not sufficient evidence for an integrity charge. This stops the lawsuit pipeline immediately. - Phase 2 (Month 3–4): Stand up a redesign fund — ~$3K–$8K stipends per faculty member for course-level assessment overhaul, prioritizing high-enrollment gateway courses. - Phase 3 (next assessment cycle): Department-level reporting on assessment redesigns completed and integrity-case volume.

Required resources: $150K–$400K depending on institution size; 0.25 FTE in the teaching center to coordinate. Success metrics: decline in formal integrity cases adjudicated on detector output; faculty self-reported confidence in assessment validity; absence of new detector-based litigation. Risk mitigation: faculty in large-enrollment courses will (correctly) push back that authentic assessment doesn’t scale at 300:1 ratios — pair this with TA support, not just exhortation.

3. Treat the entry-level job collapse as an enrollment and curriculum signal, not a careers-office problem.

The Yale CELI / Sonnenfeld analysis that AI is removing the first rung of the career ladder rather than the senior rungs AI won’t kill your job — it will kill the path to your first one has direct implications for institutions whose value proposition is “graduate, get hired.” If agentic systems are absorbing the analyst, paralegal, junior-developer, and entry-marketing roles that historically converted a bachelor’s into a salary, the ROI argument for several large majors weakens at exactly the moment institutions face the demographic cliff. Recent policy literature is already framing AI deployment as a retention-and-survival response to enrollment crisis Risk, Retention, and the Algorithmic Institution — which is a worrying frame because it makes pedagogical decisions a function of solvency.

Recommended alternative: commission a credit-hour-level analysis of which majors’ employment pipelines are most exposed, and front-load curricular redesign there — not as “AI literacy modules” but as restructured capstones, applied projects, and employer co-designed sequences that build the judgment AI can’t yet perform L’IA sait tout produire… mais pas encore juger.

Implementation framework: - Phase 1 (Month 1–2): Provost-led labor-market exposure audit by program. - Phase 2 (Month 3–6): Targeted curriculum-renewal grants to the three to five most exposed majors. - Phase 3 (one-year horizon): Reported placement and graduate-school yield by program, with year-over-year comparison.

Required resources: institutional research analyst time; ~$50K–$150K renewal fund. Success metrics: changes in first-destination outcomes; employer partnerships per program. Risk mitigation: don’t let this become a pretext to cut humanities — judgment is the surviving good, and humanities teach it.

4. Build a student-voice mechanism with actual veto points, not a survey.

When the University of Staffordshire ran a course “taught by AI,” the students who paid tuition pushed back hard and publicly We could have asked ChatGPT’: students fight back over course taught by AI. The pattern is consistent across the Cal State and ASU stories: students discover the AI integration after the fact, through the syllabus or the LMS. A student-experience survey, deployed after the rollout, is theatrical consultation.

Recommended alternative: a standing Student AI Council with formal advisory standing on procurement decisions and a published response requirement — administration must answer council objections in writing before signing.

Implementation framework: - Phase 1 (Month 1–2): Charter the council with seats reserved for graduate students, transfer students, and disabled students (groups disproportionately surveilled by detection systems and disproportionately reliant on accommodations the tools may break). - Phase 2 (Month 3–4): Council reviews current AI deployments and produces a public report. - Phase 3 (ongoing): Council pre-review of any new tool affecting >5% of enrolled students.

Required resources: stipends ($1K–$2K per student per semester); 0.25 FTE staff liaison. Success metrics: published council reports; documented administrative responses; reduction in surprise-rollout incidents. Risk mitigation: don’t let the council become a rubber stamp — preserve dissent in the public record.

5. Verify your own AI-policy citations before publication.

This sounds trivial; it isn’t. South Africa’s national AI policy was found to cite fabricated, AI-generated sources South Africa’s AI policy cited fake research, created by AI. University strategic plans, accreditation self-studies, and senate AI policy documents are being drafted with the same tools, on the same timelines, by the same overstretched staff.

Recommended alternative: a verification protocol — every citation in any institutional AI policy document is checked against the original source by a named human before the document goes to the board. Required resource: librarian time, which most institutions already have. Success metric: zero hallucinated citations in board-approved policy. The cost of getting this wrong, after South Africa, is reputational in a way no AI strategy survives.

Supporting Evidence

The Evidence Base: What Leadership Can and Cannot Conclude

Evidence Landscape

This week’s corpus draws from 6252 articles, with 2287 in the higher-education category. The evidence is uneven in rigor: peer-reviewed work on assessment redesign and writing pedagogy sits alongside vendor help-center documentation and institutional press releases announcing strategic partnerships. That asymmetry matters for any leadership team weighing strategy. The most methodologically careful pieces — a systematic review on writing assessment Reimagining Writing Assessment for the AI Era: A Systematic Review on Balancing AI Support and Authentic Skill Growth, a critical paper against AI detection in assessment Contra generative AI detection in higher education assessments, and a comparative learning-analytics study of ChatGPT versus human experts Unpacking help-seeking process through multimodal learning analytics:A comparative study of ChatGPT vs Human expert — converge on a narrow finding: tools work in narrowly-scoped, well-supervised contexts. They cannot tell you whether a campus-wide deployment will hold up across disciplines, instructor expertise levels, or four-year curricular arcs.

What the evidence does establish: detection-based enforcement is failing legally and pedagogically AI Detection Lawsuits: Every Student Case, Outcome, and What the Data …, An Adelphi University student was accused of using AI to … - Newsday. What it does not establish: that any specific vendor partnership delivers the learning gains its sales materials imply.

Stakeholder Perspective Gaps

The corpus contains no systematically captured student or contingent-faculty voice on the deals being signed in their name. Coverage of the Cal State–OpenAI partnership surfaces this only because students and faculty publicly refused to use it Cal State struck a deal with OpenAI. Some students and …; the Staffordshire course-taught-by-AI story surfaced through student complaint 'We could have asked ChatGPT': students fight back over course taught by AI. When the people most affected only enter the record through resistance, your governance process is not collecting their input — it is waiting for their lawsuits.

Documented Failure Patterns

Three failure categories appear repeatedly. First, governance failures: South Africa’s national AI policy cited fabricated, AI-generated research South Africa's AI policy cited fake research, created by AI — a cautionary case for any provost drafting strategy under deadline pressure. Second, due-process failures: the Adelphi litigation Adelphi University accused a student of using AI to plagiarize. He … and the broader lawsuit pattern document institutions losing on detector-based accusations. Third, deployment failures: the ASU course-builder rollout drew faculty objections about pedagogical authority being routed around shared governance Faculty Concerned About ASU's New AI Course Builder.

The pattern across all three: institutions moved faster than their evidentiary, procedural, or consultative infrastructure could support.

Power and Framing Analysis

The dominant frame in this week’s coverage is the “embed AI in every degree” announcement — Surrey is the cleanest example Surrey embeds AI in every degree from 2026. This framing positions the university as the actor and AI as a neutral capability. It obscures who actually controls the substrate: OpenAI’s ChatGPT Edu terms ChatGPT Edu at OpenAI - OpenAI Help Center, the persuasion dynamics MIT Sloan documents in commercial models How generative AI 'persuasion bombs' users, and the algorithmic-institution logic UTP describes as a retention response to the enrollment cliff Risk, Retention, and the Algorithmic Institution: Artificial Intelligence as a Policy Response to Higher Education in Crisis. The “tool” metaphor lets a vendor’s quarterly product roadmap quietly set the terms of a two-semester curriculum.

Research Gaps Affecting Strategy

Leadership is making multi-year commitments without longitudinal evidence on graduate competence, labor-market interaction, or differential effects across student populations. The Fortune reporting on collapsing entry-level pipelines AI won't kill your job — it will kill the path to your first one raises a question the pedagogical literature cannot yet answer: are we credentialing students into a job market whose first rung has been removed?

Secondary Tensions

Beyond the headline tension between adoption and integrity, three quieter conflicts deserve naming: grading-automation ethics versus faculty workload Is It Ethical to Use AI to Grade? - Education Week; authentic-assessment redesign versus the credit-hour and accreditation cycle Beyond Detection: Redesigning Authentic Assessment in an AI … - MDPI; and governance frameworks that scale AI Leadership in Education: A Governance Framework to Scale Safely versus the local academic-freedom claims Writing with machines? Reconceptualizing student work in the age of AI that scaling tends to override. None of these resolve through better tooling.