8 Responsible Use in Education

Responsible Use Applies Throughout the Workflow

The previous chapter, Evaluating AI Outputs and Interactions, introduced ABC-R. Its first three checks examine the claim, the frame, and the fit: Accuracy and Evidence, Bias and Perspective, and Context and Relevance. It also asks how your prompts and follow-ups shaped the answer. The R asks a further question: Should this AI use happen under these conditions? That question applies throughout the work. Privacy and tool access, for example, need attention before the first prompt.

Four questions can help faculty think through those conditions. They are this training’s practical synthesis of concerns in the U.S. Department of Labor’s AI Literacy Framework, including sensitive information, workplace and legal requirements, possible harm, risks that vary by situation, and responsibility for outcomes.

Four Questions for Responsible AI Use
Question What to examine
What must remain human? The judgment, relationship, expertise, or learning the task requires.
What can the tool see or do? The data, accounts, files, permissions, and actions involved.
Who else is affected? Students, colleagues, authors, workers, communities, and people asked to trust the result.
How will the use be accountable? The policy, approval, review, documentation, disclosure, or citation the situation requires.

The questions overlap. A decision about tool access may also affect students’ privacy or their ability to take part. They can guide both our own choices and conversations with people who question those choices.

What Must Remain Human?

Start with what the task asks a person to do. Using AI to reorganize headings in your planning notes requires a different level of review than using it to recommend grades, write student feedback, interpret a policy, or identify students who may need help. Human oversight means that a person reviews the AI’s contribution, makes the decisions, and accepts responsibility for the result.

Using AI for Feedback

AI can support learning through feedback when we choose its role with care. Formative feedback helps students improve work they can still revise. A faculty member might ask AI to suggest comments, then check and adapt them based on their own reading of the work. Students can also use AI feedback directly in an activity that asks them to judge its usefulness. In either case, decide what students should learn from the feedback and give them time to act on it.

Automated grading puts this responsibility to the test. Marc Watkins describes some of the many new tools that promise to evaluate student work and generate feedback. A faculty member remains responsible for whether that evaluation is accurate, fair, aligned with the assignment, and appropriate for the student. In many cases, the educational relationship itself requires the instructor to read the work and make the judgment. Students need and expect a teacher who understands their work well enough to discuss it with them.

In PAIRR (Peer and AI Review + Reflection), students use peer and AI feedback to revise their own work. Students compare peer and AI comments on a draft, decide which suggestions fit their goals, then revise and explain their choices. The activity gives students practice judging feedback while keeping human readers involved (Sperber et al., 2025). Its developers also discuss their reasons for using AI and concerns about privacy and bias in their ethics statement. That discussion gives faculty a way to involve students in the reasons for an AI use and the concerns it raises. (Part II explores PAIRR in more detail, including how to build it into the revision process.)

Using AI for Tutoring and Practice

The same concern for learning applies when AI acts as a tutor or study partner. Students need chances to practice the abilities a course is meant to develop. AI can help them finish a problem while leaving them less prepared to solve the next one on their own. It can also guide practice in ways that support learning. Which happens depends partly on how the tool and the interaction are designed.

Two studies of AI tutors help explain this difference. In a study of nearly 1,000 high school math students, one group practiced with a version of GPT-4 that worked like a standard chatbot. Their practice scores were 48% higher than those of students who had no AI access. When they later took an exam without AI, their scores were 17% lower. A tutor version designed to give teacher-guided hints instead of answers raised practice scores by 127% and largely avoided the later drop (Bastani et al., 2025). The practice results alone would have missed the difference in what students could do without help.

In another study, students in a large introductory physics course at Harvard completed two lessons with an AI tutor or through in-class active learning, where students worked through problems together. The tutor used teaching practices such as breaking problems into steps and responding to students’ answers. Students using it made greater learning gains in less time on those lessons (Kestin et al., 2025). Both studies give us reason to examine the kind of help a tutor provides and what students still have to think through themselves.

Governance: What Can the Tool See or Do?

Once you have a role in mind for AI, consider what it would need to see or change to do that work. A tool might help organize teaching materials without needing access to student records. Matching access to the task starts with the information involved, the account you use, and the permissions you grant.

The Family Educational Rights and Privacy Act (FERPA) protects education records: records directly related to a student and maintained by an institution or a party acting for it. Grades, submissions, private feedback, and discussion posts typically qualify. Removing a name may leave enough detail to identify a student. The Department of Education’s de-identification guidance explains that details can identify someone alone or in combination. A story in a discussion post, for example, may still point to its author after the name is removed.

Before uploading student work, check whether your institution has approved the service. The U.S. Department of Education advises educators to ask their school or technology office. A college may have a contract that sets rules for handling student information. A personal account with the same company may have different terms, so the company name alone cannot tell you whether your account is approved for the task.

Even an approved account raises two separate questions: whether the company may use your conversations to improve its models, and how long it keeps them. These are different settings or policies. Turning off model training does not by itself delete stored conversations. Deleting a chat from your history may also leave other copies for a period of time. Check the terms for the account and service you actually use.

Privacy and Approval Checks Before Using AI with Student Information
Situation Responsible starting point
Names, student IDs, grades, advising details, or private feedback Do not enter them into an unapproved tool. Follow the institution’s approved process.
Student submissions, discussion posts, recordings, or class transcripts Check institutional guidance before uploading. Remove names and identifying details when required, but check whether the remaining work could still identify a student.
A hypothetical teaching example Use invented names and details that cannot be traced to an actual student.
A tool requesting access to a drive, inbox, browser, or learning management system Check approval and permission scope. Give the least access needed, use copies, and keep sensitive accounts disconnected when possible.

The same access question becomes broader with an AI agent, a tool that can carry out several steps on your behalf. As Agentic AI showed, an agent may work in a folder on your computer, inspect browser pages, connect to an account, edit files, or run on a schedule while you are away. It may reach information you never pasted into a chat. Before using one, identify what it can see, what it can change, and which actions require your approval.

Work with copies and require confirmation before sending, submitting, deleting, purchasing, or changing the only version of a file. A long series of approval requests can lead you to click yes without reading. Deciding in advance which actions need your review can make those pauses more useful. Disconnecting an agent stops that connection’s future access, but files or messages it has already copied may remain.

Your institution also has work to do here. The NIST Generative AI Risk Management Profile assigns roles across an organization for privacy, security, human review, and records. Faculty need to know where to ask about approved tools and report problems. When you cannot establish whether an account or task is approved, pause and ask the office responsible for the policy or system.

Who Else Is Affected By Your AI Use?

Knowing what a tool can access helps us see whose information is involved. We also need to consider who pays for the tool, who relies on its results, and who handles its mistakes. A stakeholder is someone affected by a decision or reasonably asked to trust the result. Students whose work is evaluated are stakeholders, as are colleagues whose materials you upload and staff asked to resolve a problem.

Identifying those people is a start. We still need to understand what they hope to gain, what they could lose, and how much say they have in the decision. A person’s role helps us identify questions to ask, even if it cannot tell us what that person believes. A student may welcome private study help while objecting to AI evaluating their work. Listening to the reason for the objection helps us decide what, if anything, should change. If a student cannot use a required AI tool, offer another way to reach the same learning goal. For example, a student practicing an interview with an AI chatbot could practice with a classmate.

In a Quebec study of about 1,200 students, parents, and educators in primary and secondary education, acceptance and trust varied by stakeholder and use. Students rated AI grading without teacher review as less fair and less helpful than grading with that review. Greater privacy increased the perceived benefit of a question-answering tool (Karran et al., 2025). Asking someone whether they support AI in general would miss those differences. For a course decision, ask what they think about the particular use and why.

Shared Materials and Shared Responsibilities

Uploading a colleague’s slides or a publisher’s test bank creates an ownership and permission question. Do you have the right to provide that material to the tool? Authors and artists may also have concerns about how their work is used. Jon Ippolito’s Recent News for AI and the Law offers an overview of recent U.S. court decisions on AI and copyright. Legal rules help set boundaries, while conversations with colleagues can clarify expectations about shared work.

Adopting an AI tool can also change other people’s responsibilities. Suppose a department introduces an AI assistant to answer students’ course questions. Someone must keep its course information current and respond to questions it cannot answer. A system that saves one person’s time may create new work for someone else. Faculty and staff who would do that work should help decide whether the arrangement is reasonable and what support it needs.

Discussing Concerns and Disagreements

A choice about classroom AI can become a wedge issue when people take it as a sign of what someone values. These choices can touch our broader beliefs about what education is for and which kinds of human work deserve protection. Someone can understand how a tool works and still object to using it (and it’s often the case that people who strongly dislike AI understand it very well!). Consider two instructors who both want students to learn. One sees help at any hour as part of the college’s duty to make learning accessible, while the other gives more weight to students working through problems on their own. They may agree on what an AI tutor can do and still differ about how much guidance a course should provide.

In conversations like this, replies such as “You just need to learn how to use it” or “AI is here to stay” dismiss the concern before we understand it. Ask what the person hopes to protect and what experiences have shaped that view. You might ask a student, “What would you want to keep in your own hands?” or a colleague, “What do you most want students to learn through this work?” Listen to the answer and check that you’ve understood before explaining your own reasons. The Constructive Dialogue Institute’s Perspectives program encourages this attention to values and shared goals. Check factual claims together, while leaving room to disagree about which uses are worth choosing.

The conversation also depends on your role. Students may worry that questioning an instructor’s AI use will affect how their work is judged. It can help to offer a private way to raise concerns and explain how they can seek review. In a faculty discussion, colleagues may have different duties, course goals, or workloads. Ask how the proposed tool would fit their courses and what support they would need.

Communities and Environmental Costs

AI’s effects on people far from the classroom can raise similar questions about whose needs and concerns should guide our choices. AI services depend on workers and data centers that use electricity and water. Communities may gain jobs and tax revenue while also facing new demands on local resources. NIST’s AI Risk Management Framework considers effects on people, organizations, communities, society, and the environment. For an institution choosing a service or signing a contract, those wider effects are part of the decision.

Students and our own colleagues may hold strong views about infrastructure related to AI, such as data centers. In a March 2026 Gallup survey, 71% of U.S. adults opposed building an AI data center in their area. Polling research suggests the reasons are highly variable, with some concerned about NDAs, undisclosed tax breaks, the aesthetics of a data center close to home, an environmental impact such as pollution or water usage. AI and data centers have become intertwined issues for many and mundane uses of AI in the classroom can spark heated discussion. Yet these discussions can be opportunities for faculty. A class discussion can examine which costs and benefits people value, who should have a say, and how AI compares with other online services students use every day. What Your Digital Life Uses, mentioned in another chapter, lets participants estimate the energy and water behind a day of their own cloud-connected activity. This kind of “digital inventory” can invite students (and colleagues) to consider AI as part of a broader discussion about our entire digital lives, including how we spend our time and allocate our attention.

AI provides enormous benefits and capabilities for faculty to support their work and students, but we should be prepared to have these types of conversations and explain our own reasons for adopting (or not) these new technologies.

How To Remain Accountable

When we choose to use AI, those affected should be able to understand that choice and question the result. AI can help us make courses more accessible and interactive, especially online. It can draft captions, help build practice activities, or respond to students’ questions while we are away. As more of that work becomes automated, however, students need to know which parts came from AI and where our judgment shaped the result. Accountability means being answerable for those choices and their effects. It includes following policy, reviewing the work, explaining our decisions, and keeping records when the stakes or policy call for them.

In Contestable AI by Design, Alfrink and colleagues argue that people affected by automated decisions need ways to understand and challenge them. Contestability means that a system is open to those challenges and can respond to them. An explanation should help someone see why a decision was made, and a reviewer must have the authority and ability to change it. The authors place these responsibilities throughout a system’s life, from its design to its use and review.

In a course, we can make that responsibility visible through the materials students use. Suppose you use AI to develop an interactive practice activity for an online lesson. Tell students which parts AI helped create and which learning goal you chose. You might show one generated question you dropped because it gave away the answer, or explain why you rewrote feedback to ask students to reconsider a step. A short note alongside the activity can point to those choices. Students can then see what you checked and how your judgment changed the result.

We also model accountability for students when we explain these choices. If we ask them to show how they used AI and what thinking remained theirs, we should be ready to do the same. When a student points out an error, explain how you corrected it and what you will check next time.

Explaining AI’s Contribution

An AI disclosure statement describes how AI contributed to a particular work product. It should tell readers what tool you used, what you asked it to do, what you checked or changed, and what judgment remained yours. Include rejected suggestions when they help explain your decisions.

For course materials, a disclosure might read:

I used Microsoft Copilot to suggest a clearer order for the headings in this study guide. I selected the final organization and wrote the explanations and examples. I checked all claims against the assigned course materials and remain responsible for the final guide.

For a committee report, it might read:

I used Claude to compare the headings in three public policy documents and create an initial comparison table. I checked the table against the original documents, corrected two classifications, and wrote the analysis and recommendations.

Citing Material Used in the Work

Disclosure and citation answer different questions. A disclosure explains AI’s role in the work. A citation points readers to material you quoted, paraphrased, analyzed, or treated as a source. Some situations require one; some require both.

Current style guidance is still developing. The MLA’s updated guidance recommends citing AI-generated material when it is quoted, paraphrased, or incorporated into a work. Its guidance on describing AI use explains why substantive use may need an acknowledgment beyond a citation. The Chicago Manual of Style often treats a statement in the text as sufficient, while allowing a note when a formal citation is needed. Follow the requirements of your institution, discipline, publisher, grant, or professional setting.

If a chatbot summarizes a source you want to use, open the original, read it, and cite it directly. Sometimes the AI output itself is what you want to discuss. In that case, give readers enough detail to understand the exchange. Include the tool, model if known, date, and prompt or task, along with a stable link or saved record when one is available.

Keeping Records When Needed

Internal documentation records enough of a higher-stakes use for someone to review it. Routine uses may require no saved record. Grading, research, purchasing, or program-level decisions may require one under institutional policy. Check those requirements before the work begins so you know what needs to be retained. A disclosure statement does not override a policy that prohibits the use.

Sarah’s example brings the four questions together. Her decision about what to share changes the task she gives AI and what she later tells students.

What to Keep

  • Evaluating AI Outputs and Interactions evaluates the output and the interaction. Responsible Use evaluates the conditions before, during, and after using AI.
  • Human oversight requires a person who can question the tool and remains responsible for the decision.
  • Check institutional approval before student information enters an AI system. Use the least data and access needed.
  • Identify people affected by the use and ask about their concerns. If you require a tool, design an accessible way to reach the same learning outcome.
  • Disclosure explains AI’s role. Citation documents AI-generated material used as a source. The context may require one or both.
  • A responsible decision can be to use AI, narrow the workflow, choose a different tool, or not use AI.

Looking Ahead

This chapter concludes Part 1. You now have a foundation for understanding AI, directing it, evaluating its outputs, and deciding whether a particular use is responsible. Part 2 turns toward course and assignment design. It examines learning outcomes, AI-use expectations, authentic assessment, visible learning processes, and accountability. The Responsible Use questions from this chapter will return whenever you decide what role AI should play in a course or assignment.

References

Alfrink, K., Keller, I., Kortuem, G., & Doorn, N. (2023). Contestable AI by design: Towards a framework. Minds and Machines, 33, 613–639. DOI record

Autio, C., Schwartz, R., Dunietz, J., et al. (2024). Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile. National Institute of Standards and Technology. DOI record

Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26), e2422633122. DOI record

Chicago Manual of Style. (n.d.). Citation, documentation of sources: Generative AI. Source link

Gamberg, J., MacArthur, M., Mills, A., Sinha, A., & Sperber, L. (2026, August 25). AI is problematic. Why does PAIRR use it? An ethics statement for the Peer & AI Review + Reflection Project. University Writing Program, UC Davis. Source link

Gallup. (2026, May 13). Americans oppose AI data centers in their area. Source link

Gladd, J. (2026b). What your digital life uses. Source link

Ippolito, J. (2026). Recent news for AI and the law (Version 1.0). New Media at the University of Maine. Source link

Karran, A. J., Charland, P., Trempe-Martineau, J., et al. (2025). Multi-stakeholder perspective on responsible artificial intelligence and acceptability in education. npj Science of Learning, 10, 44. DOI record

Kestin, G., Miller, K., Klales, A., Milbourne, T., & Ponti, G. (2025). AI tutoring outperforms in-class active learning: An RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports, 15, 17458. DOI record

Modern Language Association. (2025, August 13). Beyond citation: Describing AI use in your work. Source link

Modern Language Association. (2025, August 13). How do I cite generative AI in MLA style? Updated and revised. Source link

National Institute of Standards and Technology. (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0). DOI record

Sperber, L., MacArthur, M., Minnillo, S., Stillman, N., & Whithaus, C. (2025). Peer and AI Review + Reflection (PAIRR): A human-centered approach to formative assessment. Computers and Composition, 76, 102921. DOI record

U.S. Department of Education, Student Privacy Policy Office. (2012). Data de-identification: An overview of basic terms. Source link

U.S. Department of Education, Student Privacy Policy Office. (n.d.). Family Educational Rights and Privacy Act (FERPA). Source link

U.S. Department of Education, Student Privacy Policy Office. (n.d.). I want to use an online tool or application as part of my course. However, I am worried that it is a violation of FERPA. What should I do? Source link

U.S. Department of Labor, Employment and Training Administration. (2026, February 13). Training and Employment Notice No. 07-25: The U.S. Department of Labor’s Artificial Intelligence Literacy Framework. Source link

Watkins, M. (2025, October 10). The dangers of using AI to grade. Rhetorica. Source link

Constructive Dialogue Institute. (n.d.). Perspectives. Program description

Further Reading

  • Eaton, L. (compiler). (n.d.). Syllabi Policies for Generative AI repository. A collection of faculty AI policies and disclosure approaches.
  • Jackson State University. (n.d.). AI syllabus policy resources. An institutional landing page for AI policy examples and related resources.
  • Leon Furze’s Teaching AI Ethics project offers further resources on privacy, data, copyright, labor, and power.
  • Fernandes, D., Villa, S., Nicholls, S., et al. (2026). AI makes you smarter but none the wiser: The disconnect between performance and metacognition. Computers in Human Behavior, 175, 108779. DOI record
  • Gladd, J. (Ed.). (2026a). The systems behind AI. In CWI 101. College of Western Idaho. Source link
  • Gladd, J. (Ed.). (2026c). Who else is affected? In CWI 101. College of Western Idaho. Source link
  • Meta. (2026, September 3). Kuna, we are online! Source link
  • U.S. Department of Education, Office for Civil Rights. (n.d.). Disability discrimination: Technology accessibility. Page last reviewed January 14, 2025. Source link

License

Icon for the Creative Commons Attribution-NonCommercial 4.0 International License

A Guide to Teaching and Learning with Artificial Intelligence Copyright © by Jason Blomquist; Liza Long; and Joel Gladd is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License, except where otherwise noted.