AI anxiety is a specific, understandable form of workplace stress: unease about job security, ethical use, or personal competence when a new technology is introduced. It is common. EY's (2023) survey of US employees found 71% were concerned about AI, and almost half said their concern had grown over the previous year, driven less by the technology itself than by a lack of clear guidance from leaders on how it should be used responsibly. Anxiety rarely shows up as outright refusal. It appears as disengagement, over-caution, or unspoken doubt that never reaches a leader unless it is actively invited in.
Risk aversion is a related but distinct response. Kotter and Schlesinger (2008) identify four reasons people resist a change: self-interest, where someone fears losing status, workload, or influence; misunderstanding and mistrust, where the reasoning has not been explained credibly; genuine disagreement, where their assessment of the risk differs reasonably from the leader's; and low tolerance for change itself, an understandable discomfort with disruption and new learning. Avoidance, quietly working around a new tool, delaying its use, or supporting it in meetings but not in practice, is often easier to spot than to name. A challenging conversation is needed to name it directly, rather than assume the hesitation is simple stubbornness.
Building genuine buy-in starts with matching the response to the real cause of resistance. Kotter and Schlesinger (2008) recommend honest communication where misunderstanding is the issue, participation where people need a stake in the solution, and facilitation and support, training, coaching, and practice time, where the gap is confidence or skill. More recent evidence supports this: employees are far more comfortable with AI when every level of the organisation is genuinely involved in shaping how it is adopted, and respond well when senior leaders visibly model using it responsibly rather than simply mandating it (EY, 2023). Involvement backed by real training, not just a policy announcement, is what measurably increases how often people choose to use a new tool (Roth, 2025).
How the conversation itself is held matters as much as its content. Edmondson (2019) argues that psychological safety, not simple comfort, is what allows people to raise real concerns: framing change as something the team is learning together, actively inviting a different view, and responding to concerns with curiosity rather than defensiveness. Guidance from Acas (2014) on challenging conversations adds a practical discipline: prepare before the conversation, focus on facts rather than assumptions, and listen for what is actually being said rather than rehearsing a reply. Stone, Patton and Heen (2010) make a related point: a difficult conversation is rarely just one conversation. It usually holds a disagreement about what is happening, a disagreement about the feelings involved, and a question about what the change means for someone's competence or identity. Separating these threads, and asking questions instead of asserting conclusions, turns a defensive exchange into a useful one.
None of this removes the discomfort of raising a hard topic. It does mean leaders can walk into it prepared: expecting anxiety and risk aversion as normal responses, naming them directly, and treating the conversation itself, not just the announcement that follows it, as the moment that builds the trust AI-ready workflows depend on (CIPD, n.d.).
Action Point
Before your next AI or automation conversation, name one person likely to resist and identify which cause, self-interest, mistrust, differing judgement, or low tolerance for change, is most likely driving it. Prepare two open questions that invite their view before you explain the decision. Hold the conversation, then note what you learned about the real cause of their resistance.