Kotter developed his eight-step model by studying change initiatives that failed. His research identified eight corresponding errors: one for each step skipped or poorly executed. Understanding these failure modes is as important as understanding the steps themselves.

For ethical AI specifically, these failure modes manifest in recognisable patterns:

Allowing too much complacency: the organisation does not genuinely believe it has an ethical AI problem. “Our AI is fine.” “We’re not doing anything harmful.” This complacency is often broken only by a crisis: a public failure, a regulatory investigation, a whistleblower. The cost of waiting for a crisis is always higher than the cost of acting without one.

Failing to create a sufficiently powerful guiding coalition: ethical AI is delegated to a small team, usually within the technology function, without executive backing or cross-functional diversity. The team produces good work that goes nowhere because it lacks organisational authority.

Lacking a clear vision: ethical AI becomes a list of principles or a compliance checklist rather than a compelling direction. People cannot connect their daily decisions to the broader purpose.

Under-communicating the vision: leadership announces ethical AI commitments publicly but fails to embed them in internal culture, decision-making, and incentive structures. The gap between external claims and internal reality generates cynicism.

Not removing obstacles: motivated individuals run up against systems, processes and cultures that make ethical AI practically impossible. They eventually give up or leave.

Failing to create short-term wins: the transformation feels endless with no visible progress. Sceptics are vindicated. Supporters lose faith.

Declaring victory too soon: early wins are mistaken for permanent change. The initiative is wound down before the new practices are truly embedded. Within months, the old equilibrium reasserts itself.

Neglecting to anchor change in culture: ethical AI remains a project, not a practice. When the project ends, so does the ethics.

Recognising these failure modes is not pessimistic. It is practical. Every organisation pursuing ethical AI transformation will encounter some of these pitfalls. The question is not whether: it is when, and whether you are prepared to recognise and respond to them.