AI Policy in Higher Education: A Practical Template

An AI policy in higher education should state when AI assistance is allowed, how people disclose it, what information stays out of tools, and who remains responsible for the work. Use an institution-wide baseline plus assignment and research instructions that make those boundaries concrete.

The template below is a starting point for a US college or university to review through its own approval process. fenbs can retain approved departmental rules and the decisions behind them, so connected assistants start with the same instructions faculty have agreed.

What needs to be decided before adopting the template?

Name a policy owner, an approving body, an effective date, and an accessible contact for questions. Bring faculty, students, disability services, the library, information security, and the research office into review. A teaching rule should not quietly become permission to upload research participants’ interviews.

Decide how institutional requirements interact with course instructions. A useful arrangement lets instructors narrow or expand academic uses within approved privacy and accessibility boundaries. Publish the default when an assignment says nothing, and tell students how to resolve conflicting directions before submitting.

For policy foundations shared with other organizations, see the AI policy guide. This template concentrates on the words faculty, staff, and students need at the point of use.

What policy language can a college adapt?

Replace every bracketed field before approval. The following clauses are proposed institutional choices, not a statement that every college is legally required to use this wording. Keep the approved version alongside existing academic integrity, accessibility, security, and research policies.

• Purpose and scope: [Institution] permits responsible AI assistance that supports learning and university work. This policy covers students, faculty, staff, and others acting for the institution, including AI features within familiar software.
• Permitted use: Use only tools and accounts approved for the proposed information and purpose. Permission to use a tool for public materials does not authorize student records, restricted research, or confidential institutional information.
• Academic work: Follow the instructions for each assignment. Where instructions are unclear, ask the instructor before using AI. Do not present generated analysis, writing, code, or media as unaided work when that would misrepresent the assessed contribution.
• Disclosure: When required, identify the tool, the kind of assistance, and the parts of the submitted work affected. Describe substantive changes you made and how you checked the result. Do not include confidential prompts or protected information in a disclosure.
• Responsibility: The submitting person remains responsible for accuracy, citations, permissions, and meeting the learning objective. Read cited sources and verify calculations. An AI response is not evidence that a source exists or supports a claim.
• Information handling: Do not enter education records, personal information, confidential examinations, unpublished restricted materials, credentials, or research data unless the appropriate institutional office has explicitly approved that use and its safeguards.
• Assessment and review: Instructors explain the permitted assistance and evidence of learning before assessment. Concerns about prohibited use follow the established academic integrity process, including an opportunity for the student to respond.
• Research: Follow sponsor, publisher, participant-consent, and institutional research requirements. The principal investigator resolves questions with the research office before restricted materials enter an AI service.
• Questions and incidents: Ask [contact] for guidance. Report suspected information exposure or materially harmful output to [incident channel] promptly. Policy exceptions require written approval from [authority] and state their purpose, limits, and end date.

Make the instructions easy to find from the syllabus, staff handbook, and approved-tool page. A policy that exists only in a committee attachment leaves people to invent their own defaults. Keep those defaults in fenbs Decisions and rules after a person approves them.

How should assignment permission and disclosure work?

Use a short assignment label that explains the boundary in ordinary language. “AI permitted” is too broad if brainstorming is acceptable but generated answers defeat the purpose of the exercise. Describe the allowed action and the evidence the instructor will assess.

Offer a disclosure sentence students can adapt: “I used [tool] to [activity]. I retained [contribution], changed [changes], and verified [checks].” State whether ordinary spelling tools need disclosure. Avoid requiring complete chat histories when a focused explanation demonstrates the relevant contribution.

Plan an accessible route for students who cannot use a required service or would need to disclose disability information to obtain help. A department should decide that route before the deadline. Do not treat a detector score alone as a substitute for reviewing evidence and hearing the student’s explanation.

What privacy and research boundaries belong in the policy?

FERPA, the federal student-records privacy law, matters when an AI workflow discloses protected information from education records. The Department of Education explains that online services need a case-specific privacy review; removing a name may leave indirect identifiers. Use its educational-services guidance↗ (opens in a new tab) when assessing the workflow.

For example, replacing a student’s name with initials in a distinctive accommodation request does not make it suitable for a public chatbot. Have the privacy office assess the data, disclosure basis, vendor terms, and access. A departmental policy cannot approve away institutional requirements.

Research needs a separate check for the particular sponsor and activity. NSF’s merit-review policy↗ (opens in a new tab) prohibits reviewers from uploading proposal content and related review records to non-approved generative AI tools. Do not extend classroom permission to peer review or assume all funders use identical rules.

How do you test the policy before a semester starts?

Walk through an instructor drafting feedback, a student checking code, a lab assistant summarizing interviews, and a staff member preparing a public event notice. For each scenario, ask who approves the tool, which inputs are allowed, what gets disclosed, and who checks the result.

If two reviewers reach different answers, revise the clause or add a concrete example. The guide to using AI at work gives staff a practical starting point; the AI brain explainer explains why keeping approved work context separate from a private chat helps continuity.

How to do this with fenbs

1. Create a free account and add a project for one department’s AI policy review.
2. Add tasks for reviewing the template with teaching, privacy, accessibility, and research representatives. Keep sensitive examples in their approved institutional systems.
3. Record the decisions people approve, their reasons, and the scope of each standing rule. Leave disputed clauses as open questions.
4. Ask a connected assistant to read the project context before preparing faculty guidance. Save progress with outstanding checks and the person needed for each answer.

fenbs is not a student information system or a replacement for formal institutional approval. Keep student records and restricted research out of the policy project. Review the template when a course design, approved service, or sponsor requirement changes; the AI governance hub collects related guidance.