Why AI adoption in universities is really a question of trust

Universities are investing heavily in artificial intelligence. Yet, a look at how staff and students are using it reveals a striking disconnect: the tools we rely on often sit entirely outside institutional provision. We believe that the challenge is not primarily technological but relational. It is about who trusts whom to use AI, under what conditions, and especially in high-stakes assessment contexts. 

Three years ago, we argued that the evolution of AI from background tool to teaching partner created a new pedagogic paradigm that included students, educators, and AI. But “Pedagogic Paradigm 4.0” omitted a critical fourth force: Institutional governance, the collective of decision-makers across academic and professional services teams.

We have now entered “Pedagogic Paradigm 5.0”, a four-party model involving students, educators, AI and the institution. Here, institutional governance acts as a gatekeeper – deciding which tools are enabled, restricted or blocked – to shape our relationships with AI. Lack of trust between parties can hamper institutional AI adoption. 

Roger C. Mayer et al’s integrative model of organisational trust helped us understand how. The model suggests that trust, in an organisational context, is influenced by ability (perceived competence of an actor within the organisation), benevolence (goodwill and concern for others) and integrity (adherence to a set of principles, such as fairness and consistency). 

• Spotlight guide: AI and assessment in higher education
• Spotlight guide: Reframe assessment as part of learning
• ‘Students don’t have to prove authorship of every word, they show their supervision of AI tools’

How this might look in a university setting:

Ability: the extent to which those responsible for AI governance perceive an educator to be competent at their job. 

Benevolence: the idea that the institution has the educator’s best interests at heart, and as such, acts in ways that support rather than undermine their success. 

Integrity: an obligation to follow institutional guiding principles around issues of security and fairness.

When institutions adopt AI, they tend to focus on integrity: risk control, data security and maintaining the status quo. This sometimes means restricting AI use or providing/allowing the use of tools with limited capabilities. In such instances, educators face a choice between following the rules or using shadow AI; in other words, AI tools that fall outside the approved list.

If an educator finds a tool that will significantly improve their teaching but their university mandates use of a different one that doesn’t perform as well, they might wonder whether the institution’s desire for safety has overridden its commitment to their success. A “benevolence gap” emerges, undermining trust. If the non-institutionally approved tool works better, and the official policy feels like a hurdle, educators start to trust the “shadow” tool more than the institutional gatekeepers that discourage its use.

So how do we work with AI transparently and meaningfully when institutional policies create constraints?

In Pedagogic Paradigm 5.0, the university’s role must evolve to ensure that students and staff can interact with AI safely and effectively.

Building trust around AI use

Be transparent: educators must start by discussing AI use openly with students, even if it is shadow AI use. We must also design assignments where the role of AI is transparent, and focus assessment on judgement, interpretation and critique rather than on first-draft production alone.

Prioritise capability: institutions must choose tools that genuinely support teaching and learning rather than merely satisfy compliance. Educators can contribute by sharing examples of real academic workflows, such as drafting, thematic coding, feedback generation or literature exploration.

Reduce friction: when rules function as barriers rather than enablers, users do not simply comply less; they redirect their trust elsewhere. Friction often appears in the form of limited access to tools, complex approval processes, inconsistent guidance or platforms that lag behind what is readily available outside the institution. These barriers signal to staff and students that using AI requires workarounds, which in turn normalises shadow practices. Reducing friction therefore means ensuring that approved tools are easily accessible and functionally competitive, and their use is supported with clear, timely guidance.

Allow for different speeds: adoption should be a choice, not a mandate. Some educators will experiment early, others will prefer cautious observation. Rather than imposing uniform rules, institutions should support different attitudes by offering training, shared examples of practice and space for honest discussions around and sandbox-style experimentation with AI tools. Such sandbox environments are standard practice in industry and regulatory contexts, and we argued for their adoption in higher education three years ago, yet progress in many universities is limited.

The new paradigm in which we are operating therefore asks universities not simply to regulate AI but to build the institutional trust conditions that allow educators and students to decide how, when and whether to rely on it.

Isabel Fischer, Susanne Beck and Joe Nandhakumar are professors in the Information Systems Management & Analytics Group at Warwick Business School.

If you would like advice and insight from academics and university staff delivered direct to your inbox each week, sign up for the Campus newsletter.