The first stage of Lewin’s model is Unfreeze: and it is perhaps the most underestimated.

Unfreezing is not about creating chaos. It is about creating readiness. Before people can change, they need to feel that change is both necessary and possible. Without this, any new practice: however well designed: will be rejected or quietly abandoned.

In the context of ethical AI, unfreezing means helping people in your organisation understand *why* the current approach to AI is insufficient. This requires honesty and courage. It means surfacing uncomfortable truths:

– That AI systems can encode and amplify existing biases
– That decisions made by algorithms are not automatically objective
– That the absence of an ethics framework is itself a choice: and one with consequences
– That the people most affected by AI systems are often those with the least power to challenge them

Unfreezing works through several mechanisms:

Creating urgency: sharing concrete examples of AI harm, regulatory changes, or reputational risks that make the case for change undeniable.

Disrupting assumptions: questioning taken-for-granted beliefs about AI. “It’s just a tool” is an assumption. “The data doesn’t lie” is an assumption. “Ethics slows us down” is an assumption. Each can be examined.

Building psychological safety: people need to feel safe enough to acknowledge problems. If raising ethical concerns is career-limiting, unfreezing will fail. Leaders must model the behaviour they want to see.

Involving people early: unfreezing is not a top-down announcement. It is a conversation. The more people feel included in diagnosing the problem, the more invested they become in solving it.

The danger of skipping this stage is significant. Many ethical AI initiatives fail not because the framework was wrong, but because the organisation was never truly unfrozen. People complied on paper and carried on as before.

Reflection question: What would it take for your organisation to genuinely acknowledge the ethical risks in its current AI practices?