Back to school this fall brings not only syllabi and lectures, but growing uncertainty about how artificial intelligence (AI) is reshaping Canadian classrooms.That uncertainty is part of a larger shift. Canada’s new AI strategy calls for broader AI literacy, stronger public trust and responsible AI adoption, including AI learning for post-secondary students.In a Canadian university case study on AI policy in higher education that I co-authored with Emily Ballantyne, acting director of teaching and learning at Mount Saint Vincent University (MSVU), we examined how faculty experience AI policy direction. We heard that faculty need practical ways to decide when AI belongs in learning, when it does not and how it changes trust, assessment and student agency.Guidance is unevenAcross higher education, guidance about AI remains uneven. Comparative studies of university policies show approaches ranging from restrictions and academic integrity rules to disclosure requirements, faculty discretion and support for responsible experimentation. Research also identifies recurring gaps in policy communication, assessment guidance, professional learning and consultation with faculty. Students may encounter different expectations across courses while instructors are often left to interpret broad institutional principles.The relationship between students and educators is at the heart of deep learning. AI has changed the conditions under which trust, learning and assessment now happen.The policy problemThere are thoughtful international AI policy frameworks for education, although many were developed before tools such as ChatGPT became widely available.Cecilia Ka Yuk Chan, a professor of education at the University of Hong Kong, developed an AI Ecological Education Policy Framework. It identifies three areas (pedagogical, governance and operational) to help universities think about teaching, institutional rules and support systems. When considering these areas in Canadian post-secondary institutions, universities also need to include equity commitments, including responsibilities to Indigenous communities, as well as inequities in teaching conditions, such as differences in employment security, workload and access to institutional support.Broad frameworks can identify concerns, but faculty need help translating them into classroom decisions. Read more: ‘Indigenizing’ universities means building relationships with nations and lands University case studyOur mixed-methods study involved 53 faculty members who completed a survey and 12 who participated in three focus groups. All were full-time or part-time faculty at MSVU and came from varied STEM-related and non-STEM areas.We first used Chan’s three areas to examine faculty views of teaching, governance and institutional support. When we analyzed survey comments and focus group discussions, concerns about trust, authenticity, fatigue and the burden of policing AI repeatedly appeared.Faculty reported limited current AI use but expected greater future use. They described concerns about academic integrity, workload, professional support and the difficulty of making fair decisions when institutional expectations were unclear.The consequences extend beyond universities. Teacher education programs prepare the people who will bring AI literacy into K-12 classrooms. Teacher candidates need clear ways to guide students, evaluate tools and protect learning.What faculty accounts told usAcross this study and the wider research, faculty appear cautious, but do not simply reject AI. Faculty often agree that students need to learn how to use AI well. Many see benefits, such as faster feedback and new forms of learning support. They also report confusion, stress and fear.Faculty described AI policies as unclear, inconsistent and top-down. Without a shared approach, instructors were left to improvise, sometimes judging students through suspicion rather than evidence. As one faculty member said: “I feel like a detective, not a teacher.”Faculty also described fatigue and the burden of deciding whether student work was AI-assisted, especially when AI use undermines trust in reflective or personal writing. One participant observed: “It undermines the development of relational skills.”This hidden work is now part of how AI is changing classroom life.Policy must account for relationshipsAI policy should address how technology changes the social and emotional conditions of teaching and learning, rather than treating trust and workload as private instructor problems. This points to the CARE Framework, which brings together four linked commitments:Critical AI literacy: Policies should support educators in designing meaningful learning and making student thinking visible. Faculty and students need support to understand AI, evaluate tools and outputs, and use it in ways that develop skill, judgment and agency.Accountable governance: Institutions need clear rules, fair processes and principled leadership for AI use. They also need accessible professional learning, technical support and regular policy review.Relational-affective pedagogy: Policies must recognize that unclear enforcement, repeated suspicion and declining trust create additional work and strain for faculty and students.Ethical orientation: Decisions about AI should remain grounded in equity, privacy, cultural responsiveness, student well-being and the human purposes of education.The first three commitments operate as mutually supporting pillars, while ethical orientation binds them together. These commitments connect institutional rules to the experiences of faculty and students. They are also consistent with Canadian educational commitments, including Indigenous perspectives that emphasize relational accountability.Universities can act in three areas.1. Recognize and support faculty workload.Faculty face additional work as they adapt to AI. They must redesign assignments, explain AI rules, address possible misconduct and support students who are also uncertain. Universities should recognize this work and provide appropriate time, training and institutional support.2. Build trust through shared AI agreements.Universities should encourage student-faculty discussions at the beginning of each course to establish clear expectations for AI use. It should be explained why only some uses are permitted. Such agreements should help students understand the purpose of the course, the role of AI and the kinds of thinking they are expected to demonstrate.3. Move to structured professional learning.Faculty, students and teacher candidates need opportunities to understand AI, evaluate tools and outputs, make sound educational decisions and examine possible consequences.This requires educators to make defensible choices about tools, learning goals, equity, privacy and student agency.Teacher education programs should be a major site for this work. Future teachers need repeated practice making AI-related decisions before they enter classrooms.Canada’s AI strategy, teacher educationUniversities and teacher education programs must decide whether their responses will rely mainly on surveillance and control or on evidence, trust and shared responsibility. Read more: We asked teachers about their experiences with AI in the classroom — here’s what they said AI literacy needs to become part of how teachers learn to plan, teach, assess and reflect. Teacher candidates should leave their programs able to ask practical questions: What does this tool do or miss? Who benefits? Who may be harmed? What learning goal is being served?Canada’s AI strategy will reach students through teachers, teacher educators, assignments, practicum experiences and assessment routines. Without a relational and practical approach, Canada’s AI literacy goals may remain broad aspirations. Building on the findings of this study, members of MSVU’s Faculty of Education are developing evidence-informed approaches to help educators make defensible decisions about AI literacy, classroom use and assessment. This work responds to faculty concerns, and is intended primarily to support teacher education and professional learning, while contributing to broader discussions about responsible AI policy and practice in higher education.Johanathan Woodworth does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.