Pillar 3Governance Produces: Policy or Guidance Draft
AI Policies & Guidelines
Translate values into operating rules that survive the next model release.
Aligns withNIST · GovernFERPAEU AI ActProcurement & data-classification policy
The idea
Policies and guidelines are where principles become enforceable. “Policy” covers five instruments with different binding force, from binding policies and standards to advisory guidelines and enabling mechanisms. Match the instrument to the decision: a rigid rule where guidance fits constrains practice, and guidance where a rule is needed leaves a gap. Write around behavior, data, and risk, not products, so rules outlast the tools they govern.
Most institutions aren't starting from scratch. Where AI is a new input to a process that already has rules (procurement, data handling, records, security), extend those policies. Write new AI-specific guidance only for questions nothing answers yet: disclosure, academic integrity, when a human must stay in the loop, and accountability when it goes wrong.
Why it matters
Policy answers the question every faculty member, student, and staffer is already asking: 'Am I allowed to do this?' Answered well, it ends the paralysis and the shadow AI that fill the vacuum. Answered badly or not at all, people freeze or improvise, and the institution carries the risk either way.
Each practice is something you can build, assign, and evidence, with a first move, an owner, and what proof looks like. Move one practice up one level at a time.
The five levels: 1 Initial · 2 Emerging · 3 Defined · 4 Managed · 5 Optimizing. Each practice describes levels 1, 3, and 5. Level 2, Emerging: some level-3 behavior exists in pockets, but informally. Level 4, Managed: level-3 behavior is applied consistently, measured, and reviewed on a set cadence.
1
AI acceptable-use guidance
Practical, audience-specific rules for AI use with institutional data and systems.
1InitialEvery course is different and most are silent; students navigate contradictory unwritten rules.
2EmergingGood practice exists in pockets, but it is informal and has not been institutionalized.
3DefinedA campus framework exists and syllabi are required to include an aligned AI statement; AI-supported online courses are checked for regular and substantive interaction (RSI) and accessibility.
4ManagedThe practice is applied consistently, measured, and reviewed on a defined cadence.
5OptimizingCompliance is tracked, students report clarity, and the framework is reviewed annually with the senate.
Next moves
Adopt a simple tier framework with model syllabus language for each tier.
Require an AI statement in every syllabus through curriculum committee channels.
Review the framework annually with faculty senate using student feedback.
Add two course-design checks: AI tutors, feedback bots, or automated grading in online courses must not replace instructor-initiated regular and substantive interaction, and every AI tool students are required to use must meet accessibility standards.
Who owns it
Provost · Faculty senate · Center for teaching · Distance education · Accessibility services
Evidence it exists
Tier framework
Syllabus template language
Annual review minutes
RSI and accessibility review checklist
3
Policy review, exception & update mechanisms
Policies are built to survive tool churn: review cycles, exception paths, version control.
1InitialIP and authorship questions about AI-assisted work are settled case by case, or not at all; review committees have no guidance.
2EmergingGood practice exists in pockets, but it is informal and has not been institutionalized.
3DefinedThe IP policy addresses AI-assisted works and institutional AI outputs; authorship and disclosure norms are published; P&T guidance says how AI use is documented and weighed.
4ManagedThe practice is applied consistently, measured, and reviewed on a defined cadence.
5OptimizingNorms are reviewed with the senate on a cycle, applied consistently across review committees, and updated as law and publisher rules change.
Next moves
Update the IP policy to cover AI-assisted works, prompts and outputs created with institutional tools, and course materials.
Publish authorship and disclosure norms for AI-assisted scholarship, teaching materials, and creative work.
Work with the senate to add AI guidance to promotion, tenure, and reappointment criteria: how candidates document AI use, and how reviewers weigh it.
Who owns it
Provost · General counsel · Faculty senate · Research office
Evidence it exists
Updated IP policy
Authorship and disclosure norms
P&T guidance on AI
In practice
Worked example
A policy that outlives the tool
An institution writes its AI policy around one named chatbot. Eighteen months later the product is gone and three successors exist; the policy is meaningless. Rebuilt under this pillar, it governs by data and behavior ('no Level-3 student data in any unapproved external AI service'), with an exception path and annual review. When the next tool arrives, nothing needs rewriting: it's classified and slotted into existing rules.
Watch for
Tool-specific rules that expire as fast as the vendor market changes.
One long legal document instead of short, audience-specific versions people will read.
No exception path, so the policy is ignored or routed around.
A freestanding AI policy that ignores existing security, privacy, and procurement rules, creating contradictions and an orphaned document no office owns.
Principles for durable policy
Two moves keep AI rules useful as technology shifts: build on the policy you already have, and govern technology-neutrally so rules outlive any tool.
Build on what you already have
Where AI is just a new input to a process that already has rules (security, privacy, procurement, records, academic integrity), extend that policy instead of writing a parallel AI rulebook that will eventually contradict it.
DoAmend the policy that already governs the process.
NotStand up a separate AI policy for a question existing policy already answers.
Govern the behavior, not the product
Write around data, behavior, and risk, never a tool or vendor. Durable language such as “AI and emerging technologies” survives the next product cycle; a policy built around one named tool dies with it.
DoFrame rules by data sensitivity, behavior, and risk tier.
NotName specific products or vendors in the binding text.
Right-size the carve-outs
Technology-neutral isn't one-size-fits-all. Some capabilities, agentic AI above all, raise questions general language can't cover: action boundaries, human-in-the-loop, auditability, accountability. Add targeted provisions without fragmenting the rest.
DoAdd focused provisions where a capability demands them.
NotLet one high-risk capability force a separate rulebook for everything.
Write only the new, and build in renewal
Write new AI guidance only for questions no policy answers: disclosure, integrity, human oversight, accountability. Give every policy an owner, a review cadence, and an exception path so it keeps pace with the technology.
DoGive each policy an owner, a review cycle, and an exception path.
NotTreat AI as purely an extension and leave the new questions unanswered.
Policies and guidelines are not the same thing
“Policy” covers five instruments with different binding force. A binding policy is not an advisory guideline; treating them alike over-controls low-risk use or under-protects high-risk use. Name the type, and let each move at its own speed: policy slowly through shared governance, guidelines and enablement quickly.
Before any of these, the instruments answer to a layer above them
Laws & regulationsBinding floor
Federal (FERPA, HIPAA, ADA / Section 508), state (e.g., Colorado AI Act, NYC Local Law 144), and extraterritorial (the EU AI Act, if you enroll EU students or operate there). Policy can exceed these but never fall below them.
Recognized standards & frameworksAdopted by choice
NIST AI RMF, ISO/IEC 42001, OECD AI Principles. Voluntary references, not binding until your institution adopts them, then they flow into your own standards.
Mission, values & principlesThe internal “why”
Pillars 1 and 2: the institutional purpose and commitments every instrument has to serve.
Laws sit above the five instruments as the binding floor. Policy translates law into operating rules, which is why this pillar says to amend existing security, privacy, and procurement policy: it already encodes those obligations. Compliance is operated in Pillar 4, Governance & Risk.
Policies
Binding requirements, with enforcement behind them.
BindingMust comply
SlowChange deliberately, through shared governance.
Standards
Measurable requirements that make a policy testable. Internal standards bind once adopted; external ones (NIST, ISO) are references until you adopt them.
Binding once adoptedMust comply
MediumUpdated as practice and technology settle.
Guidelines
Advisory, recommended approaches, not mandates.
AdvisoryShould follow
FastAdapt quickly as tools evolve.
Enablement mechanisms
Innovation support: sandboxes, approved-tool lists, incentives, and training.
EnablingMay use
FastExpand as capacity and trust grow.
Governance mechanisms
The structural processes, bodies, and reviews that keep all the rest current.
StructuralRuns the system
OngoingOperate on a standing cadence.
These five instruments run across all nine AI governance domains, from teaching to procurement, so the same vocabulary describes a syllabus rule and a vendor standard. Two tests show whether an instrument works: is it calibrated to risk (a writing assistant shouldn't face the scrutiny of a financial-aid algorithm), and can people find it? A policy no one can locate in five minutes isn't governing.
Acceptable-use grid
A starting Red / Yellow / Green map of common AI uses by audience: an at-a-glance answer to “Am I allowed to do this?” Adapt every cell to your rules before publishing.
Generally OK, Allowed with good judgment and disclosure where relevant.Caution, Allowed only with conditions, review, disclosure, or no sensitive data.Not without approval, Prohibited or requires formal governance review first.
Use case
Students
Faculty
Staff
Researchers
Administrators
Drafting routine email & internal textLow-stakes productivity; verify before sending and avoid sensitive content.
G
G
G
G
G
Summarizing public readings & documentsFine for non-sensitive material; check accuracy of summaries.
G
G
G
G
G
Brainstorming & outlining ideasIdea generation carries little risk; the human stays the author.
G
G
G
G
G
Coursework & assignmentsFor students, governed by each course's stated AI policy, disclosure usually required.
Y
G
G
G
G
Generating or drafting assessment itemsFaculty may draft with review; students may not create their own graded items.
R
Y
Y
G
Y
Analyzing data with sensitive/PII contentOnly in approved systems with the right data classification, never public tools.
R
Y
Y
Y
Y
Student records / FERPA-protected dataRequires an approved, contracted environment and data-governance review.
R
R
Y
Y
Y
Decisions on grades, admission, aid, or conductHigh-stakes automated decisions require governance review and human authority.
R
R
R
R
R
Public-facing content in the institution's nameBrand and accuracy risk, requires review and disclosure of AI involvement.
R
Y
Y
Y
Y
Syllabus statement bank
Syllabus language to adapt. Pick the stance that fits the assignment, then tailor it.
You are welcome to use generative AI tools (such as chatbots and writing assistants) to support your learning in this course, for brainstorming, feedback, and revision. When you do, include a brief note describing which tool you used and how. You remain responsible for the accuracy, integrity, and originality of everything you submit.
Readiness & maturity questions
Readiness asks: are we prepared? Use these to surface blind spots before you build. Every “no” shows where the work is. Representative prompts for reflection, not a scored test. For the scored version, use the Maturity Assessment.
Do we have institution-wide AI policies?
Are they role-specific (student, staff, faculty)?
Do they cover specific AI types (generative, predictive)?
Is there training on interpreting the policies?
Are the policies integrated into student conduct codes?
Is there a process for updating or revising policies?
Are policies designed to support innovation as well as protection?
Are ambiguities or edge cases considered (e.g., tutoring bots)?
Are guidelines adapted for classroom vs. administrative AI?
Are student rights and responsibilities clearly explained?