Designing for Virtuous AI Use: The AI-Use Ethics Matrix in AI-Mediated Classrooms

Synthesis: Petricini and Zipf argue that the ethical problem in AI-mediated classrooms is a design problem, not a compliance problem. From prior empirical work and 30 student interviews they build the AI-Use Ethics Matrix, plotting students' intention and effort against the clarity and support of institutional guidance to yield four quadrants: Virtuous Tool Use, Anxious Compliance, Opportunistic Shortcuts, and Efficient Circumvention. Their central claim is that ethical Generative AI use is an emergent property of learning environments rather than a trait of individual students: intention without clarity produces uncertainty and risk, while clarity without intention produces compliance without learning. Prohibition and detection fail, they argue, because those responses judge outputs rather than intentions and because surveillance communicates distrust, pushing legitimate AI use underground. The matrix is offered as a generative design tool and translated into three interlocking principles — explicit, pedagogically grounded guidance; process-centered assignments; and conditions that cultivate student Learner Agency.

Key Findings

1. The matrix plots two axes, not two types of student. Axis 1 is students' intention and effort toward learning; Axis 2 is clarity and support in the learning environment — internal and external conditions of ethical engagement, not policy or tool categories.
2. Virtuous Tool Use, high on both axes, is the target condition. There AI serves as tutor, evaluator, brainstorming partner or feedback mechanism; students revise output iteratively, disclose appropriately, and report confidence rather than fear.
3. Anxious Compliance was the most populated quadrant in the interview data. Students genuinely trying to learn hide legitimate uses — grammar support, concept explanations, organizing their own ideas — to avoid false accusation; its emotional signature is fear, guilt and vigilance.
4. Opportunistic Shortcuts reflect weak policy design, not moral failure. Where rules are absent and learning intention is low, students rationalize substitution ("no rule said I couldn't use AI") — though interview subjects describing these behaviors were "frequently describing peers rather than themselves."
5. Efficient Circumvention is the risk of raising clarity alone. Where rules are explicit but Assessment rewards speed and product, policy-aware students launder AI output; guidance without assessment redesign shifts students from Q3 to Q4 rather than into Q1.
6. The evidence is qualitatively grounded and mixed. The quadrants came from 30 short structured interviews at a large, research-intensive mid-Atlantic institution, coded by lumping and open coding, and the framework is also synthesized from scholarship on integrity, AI ethics and instructional design.

Why policy silence and detection both backfire

The authors frame the post-ChatGPT moment as a communication crisis rather than a technological one. Students described a "policy vacuum" in which almost no instructor had given meaningful guidance about how AI related to their learning, and silence was often read as permission. Absence of clarity does not neutralize AI use but displaces it, leaving students to rely on their own ethical leanings in isolation. Detection compounds this: it identifies outputs rather than engaging intentions, and reproduces surveillance in place of Trust and dialogue. They cite Giray et al.'s call for students to be treated as "ethical contributors rather than presumptive violators" (p. 57), and prior findings that students given room to express their own values show more concern about AI ethics than faculty and staff do.

How the matrix is built

The matrix is derived from multiple studies and refined against prior scholarship, with its two axes identified inductively rather than imposed. The authors call them the "operative variables in the ethics of AI use": the framework does not categorize students or tools but models the conditions under which forms of engagement become likely. Intention and effort show up in process artifacts such as drafts, annotations and error analyses, in metacognitive language explaining tool use, in time-on-task that places AI after independent effort, and in integrated authorship. Clarity and support show up in syllabi that distinguish permitted, discouraged and prohibited uses with concrete examples, in assessment built to accommodate transparent AI use, and in policies aligned across course sections.

Using the matrix in course design

Each quadrant implies a design intervention. Against Q4 dynamics the authors recommend oral defenses in which students explain and extend submitted work, iterative structures collecting drafts and revision notes, and prompts asking students to justify their tool choices — none of which prohibit AI, but all of which make circumvention visible and costly. They propose process-based credit for drafts, outlines or revision histories; an AI Use and Disclosure Statements log documenting when, how and why AI was used relative to the student's own thinking; and comparison tasks contrasting AI-generated with human-generated responses. They acknowledge implementation constraints — large enrollments, limited grading time, rigid curricula, uneven support — and suggest lighter-touch adaptations such as a single reflective checkpoint or a short disclosure prompt.

What this means for practice

• Treat AI policy as a relational act: state which uses are permitted, encouraged, discouraged or prohibited, explain the reasoning, and restate it in assignment descriptions and class discussion.
• Design assessment so process is visible and gradable, since students' own ethical distinction tracks whether assignments reward process or product.
• Address both axes together; guidance without assessment redesign can produce more sophisticated evasion rather than virtuous use.
• Move from detection toward disclosure-based frameworks, which the authors frame as justice-informed because surveillance concentrates false-accusation risk.
• Use the matrix diagnostically to find which axis a course is actually low on rather than defaulting to a rule change.

Limitations

• The two axes do not claim to exhaust all relevant variables; course level, disciplinary culture, student prior knowledge, and institutional context also shape engagement, and the authors invite future elaboration.
• The empirical grounding is 30 interviews at one large, research-intensive mid-Atlantic institution, coded qualitatively: the quadrants are descriptive patterns, not a validated measurement.
• The link between instructor framing and student agency is called by the authors themselves "a promising design hypothesis that warrants further applied research and practice," and several claims — detection as technocentric culture, surveillance replaced by structure — are normative positions rather than evidence.

Connected Concepts

• Academic Integrity
• AI Literacy
• AI Use and Disclosure Statements
• Trust Calibration
• Assessment
• Learner Agency

Connected Articles

• Beyond Detection: How Students Use—and Hide—AI in Online Assessments and What Authentic Tasks Can Do About It — students hide AI use in online assessments; authentic tasks respond where detection cannot.
• Purpose Before Policy: Academic Integrity, Generative AI, and Rhetorical Stance — reframing integrity around purpose and rhetorical stance instead of enforcement.
• A Guided Inquiry Approach to Students Co-Designing Generative AI Course Policies — students co-designing generative AI course policies, a classroom route to the clarity axis.

Citation

Petricini, T., & Zipf, S. (2026). Designing for Virtuous AI Use: The AI-Use Ethics Matrix in AI-Mediated Classrooms. Journal of Instructional Design and Technology, 1(2), 28-36.