Talk to UsResponsible AI resourceMenuHomeModel policyDecisionsResearch & approachGet support
Leadership guide/Research & approach
Understand the recurring provisions, the choices that vary and the approach behind our recommendations. Use the evidence to inform a policy your institution can own.
Our work began with an international discovery collection, expanded through targeted institutional research and source checks. A closer comparison of selected documents then informed the policy recommendations.
The public source library brings useful examples into view. It is a curated selection, not the full discovery collection or audit record.
828institutional discovery recordsImported University AI Policy Tracker records used to organize discovery, supplemented by targeted US research.
3,179source URLs checked for integrityAvailability, stored-text and duplication checks across the expanded collection.
5,476imported claims checked for traceabilityAutomated checks for supporting excerpts. Unresolved claims remained unresolved.
These counts describe discovery and automated source checks. They are distinct from full policy reviews and do not, by themselves, establish agreement between institutions. See how the stages connect.
These comparisons connect recurring provisions to selected official examples, meaningful differences and decisions your team needs to make. The links illustrate the findings; they are not an exhaustive list of sources examined. Open a topic to see the evidence.
Protect institutional information.
Policies repeatedly distinguish useful AI assistance from permission to expose protected information. The practical rule depends on the information, the tool and the approved use.
Evidence detail and review scope
The question is coded in 97 documents: 48 express a requirement, commitment or restriction; 6 a recommendation; 18 a conditional or local rule. It was not established in 25 examined documents.
Different scope, wording and force remain important. The count does not measure identical wording, implementation quality or the share of institutions with this practice.
Where the approaches differNorthcentral Technical College prohibits confidential, sensitive or identifiable inputs. Seminole State provides a route through approved tools, contracts and security measures. College of Western Idaho distinguishes data classes and permitted environments.
Decide which information may enter which environments, who authorizes an exception and how contracts, retention and vendor reuse are checked.
Keep people accountable.
Human responsibility and output review recur across academic and operational policies. Several sources make consequential decision authority more explicit.
Evidence detail and review scope
The question is coded in 97 documents: 65 express a requirement, commitment or restriction; 7 a recommendation; 1 a conditional or local rule. It was not established in 24 examined documents.
Different scope, wording and force remain important. The count does not measure identical wording, implementation quality or the share of institutions with this practice.
Where the approaches differJamestown bars fully automated decisions without meaningful human involvement. Dodge City names admissions, hiring and grading. Tarrant requires an authorized reviewer who can modify or reject consequential recommendations.
Name the responsible role, the decisions requiring review and the authority to intervene. Establish how affected people can seek correction.
Make meaningful AI involvement visible.
Disclosure and attribution recur, but their triggers differ substantially. A broad commitment to transparency does not settle when a label is needed.
Evidence detail and review scope
The question is coded in 97 documents: 45 express a requirement, commitment or restriction; 14 a recommendation; 18 a conditional or local rule. It was not established in 20 examined documents.
Different scope, wording and force remain important. The count does not measure identical wording, implementation quality or the share of institutions with this practice.
Where the approaches differSeminole State uses broad content-creation disclosure. Montgomery connects disclosure to substantial contribution and authorship clarity. Northcentral does not require employee content-creation disclosure, while requiring notice of autonomous interaction.
Set expectations for coursework, public communications, synthetic media, service interactions and consequential decisions. Explain routine-assistance exceptions.
Explain the rules for learning.
Course and assignment expectations are a recurring way to connect AI permissions to learning. Faculty judgment operates within institutional rules and academic responsibilities.
Evidence detail and review scope
The question is coded in 97 documents: 22 express a requirement, commitment or restriction; 9 a recommendation; 25 a conditional or local rule. It was not established in 41 examined documents.
Different scope, wording and force remain important. The count does not measure identical wording, implementation quality or the share of institutions with this practice.
Where the approaches differDartmouth undergraduate rules prohibit submitted-coursework AI use by default unless faculty permit it. Tarrant distinguishes assistive features from generative content creation. Carteret specifies course and assignment tiers.
Agree the default when a syllabus is silent, the scope of faculty discretion, accessible alternatives and how learners demonstrate the intended capability.
Support people in using AI responsibly.
Training and support appear in policies across colleges, systems and universities. The commitment may be a resource, an ongoing program or a condition of use.
Evidence detail and review scope
The question is coded in 97 documents: 42 express a requirement, commitment or restriction; 3 a recommendation; 0 a conditional or local rule. It was not established in 52 examined documents.
Different scope, wording and force remain important. The count does not measure identical wording, implementation quality or the share of institutions with this practice.
Where the approaches differCollege of Western Idaho requires annual employee training. Central Piedmont requires training before work use and refreshers. Colorado’s system procedure commits to ongoing training and leadership support.
Define preparation for each role, provide time and support, and decide what evidence will show that people can carry out their responsibilities.
Protect access to learning and service.
The reviewed sources address access through different provisions: disability accommodations, supported tools, affordability, training and human help.
Evidence detail and review scope
The question is coded in 97 documents: 17 express a requirement, commitment or restriction; 10 a recommendation; 2 a conditional or local rule. It was not established in 68 examined documents.
Different scope, wording and force remain important. The count does not measure identical wording, implementation quality or the share of institutions with this practice.
Where the approaches differDanville says students should not have to buy premium tools. Linn-Benton prohibits replacing approved accommodations with AI. College of Western Idaho requires an accessible route to a human in AI-delivered services.
Set the practical alternative when a person cannot use the tool, afford access or obtain a reliable answer. Test the experience with the people affected.
Assign ownership and keep the policy current.
Named responsibility and review arrangements connect policy to institutional governance. The decision-making structure and review schedule remain local.
Evidence detail and review scope
The question is coded in 97 documents: 54 express a requirement, commitment or restriction; 6 a recommendation; 0 a conditional or local rule. It was not established in 37 examined documents.
Different scope, wording and force remain important. The count does not measure identical wording, implementation quality or the share of institutions with this practice.
Where the approaches differMid-State specifies annual review; Louisiana’s system specifies two-year or earlier review. Piedmont uses a three-year cycle with earlier triggers. College of Western Idaho expressly keeps its AI committee advisory.
Fit responsibility to existing governance. Set a review schedule and triggers for changed tools, laws, information access, authority or material incidents.
Define authority when AI takes action.
A smaller set of reviewed documents addresses action boundaries, intervention or recovery. These provisions are especially useful when an AI system can act across connected services.
Evidence detail and review scope
The question is coded in 97 documents: 7 express a requirement, commitment or restriction; 1 a recommendation; 0 a conditional or local rule. It was not established in 89 examined documents.
Different scope, wording and force remain important. The count does not measure identical wording, implementation quality or the share of institutions with this practice.
Where the approaches differStanford communications guidance requires approval of audience-visible agent actions. Kirtland requires review before outgoing communications. Tarrant permits approved automation within predefined parameters with review, override and appeal.
Specify what a system may access, change, send or commit; when it needs approval; and how people stop, correct and recover its actions.
A common purpose leaves real choices about permissions, decision authority, faculty discretion, disclosure, access and review. The comparisons above identify those differences and link to the relevant decision guidance.
Agree what fits, adapt what needs local judgment and record who is accountable for putting the choice into practice.
Adapt the policy to your institution→
A policy can express a principle without defining the controls, resources or decision rights needed to apply it. A missing provision may also be addressed in another institutional procedure.
Our recommendations identify practical questions to test. They are not claims that every peer lacks the practice.
Ten opportunities to strengthen a policy→Review coverage and implementation→
Useful AI policy connects safeguards, institutional purpose and the capabilities people need. These principles guide our recommendations; each institution decides how to apply them.
Judge AI use by its effects on people, learning and opportunity. Keep human judgment, dignity, access and accountability central.
Give worthwhile uses a supported route to a bounded trial. Prepare learners and employees to direct, evaluate and work responsibly with AI.
Use evidence to improve, expand, pause or retire a use. Reassess information access, connected actions and responsibility as capabilities change.
For higher education, this includes teaching and learning, the academic enterprise, administration and regional workforce needs. Capability building belongs alongside privacy, security and oversight in a responsible approach.
Preparation for each role→The institutional enterprise→Systems that take action→
We organize official sources, compare provisions and distinguish what institutions say from our recommendations. Open the method or coverage notes for the detail.
The structured review produces 1,746 document-by-question assessments. Each finding is tied to the reviewed source, and each recommendation remains distinguishable from what the institution itself says.
Data inform the starting position. They do not vote on what your institution must adopt.
76US community & technical college source locations
33other US college, university & system source locations
26international source locations for comparison
The library spans 109 named institutions and systems, including 57 US community and technical colleges or systems. It includes institutional policies, implementing procedures, academic guidance and specialist documents. Multiple documents from one institution are kept distinct and identified.
International sources offer additional perspectives. Their legal context and governance arrangements do not automatically transfer to a US campus.
Explore the annotated source library→
We link to a source because it helps explain a recurring approach, a meaningful alternative or a useful operating detail. The accompanying note tells you what to look for.
Inclusion does not rank the institution, endorse its entire policy or establish that it is statistically closest to a national standard. The value is seeing how a specific decision has been expressed and what still needs your judgment.
The research begins with 828 institutional discovery records from an imported University AI Policy Tracker snapshot, supplemented by targeted official-source research. Separate collection checks examined 3,179 source URLs and 5,476 imported claims for availability, duplication or excerpt traceability. These are mechanical audit units, not additional completed policy reviews.
This edition publishes 135 official source locations: 97 structured document reviews and 38 source notes. Source notes establish relevance and scope; they are not included in the question-level counts. The September refresh retrieved text at 119 locations. Unresolved automated refreshes retain their earlier source-check status in the library.
Research coverage and review counts
828institutional discovery recordsImported University AI Policy Tracker records used to organize discovery, supplemented by targeted US research.
3,179source URLs checked for integrityAvailability, stored-text and duplication checks across the expanded collection.
5,476imported claims checked for traceabilityAutomated checks for supporting excerpts. Unresolved claims remained unresolved.
September 2026 edition. Discovery records, source locations and coded reviews describe different stages of the work. Read the coverage and method.
The comparisons are descriptive findings within a selected, mixed-scope collection. Documents from one institution and system material are not independent observations of sector agreement. No sector-wide overlap percentage, institution ranking or census of all public policies is established.
Review protocol and quality boundaries
The 18 questions cover accountability, protected data, approval, disclosure, course rules, training, access, reporting, ownership/review, consequential decisions, human recourse, learner understanding, action boundaries, outcome evaluation, records, employer validation, independent/assisted competence and role-based authorization.
Codes distinguish requirements or commitments, restrictions, recommendations, conditional/local rules, matters not established in the examined document and unresolved evidence. A silent document may be narrow in scope or rely on another procedure.
The structured reviews were conducted with a single AI reviewer and have not received independent human or second-coder validation. The counts can support inspection of this collection; they are not a validated institutional benchmark or evidence that a written policy works in practice.
Reviewed snapshots and detailed working records remain internal. Public source notes, comparison examples and the basis for recommendations are available here. Check the current official source and local applicability before adoption.