The third stage of Lewin’s model is Refreeze: and it is the one most frequently neglected.

Refreezing means embedding the new practices so deeply into the organisation’s culture, systems, and incentives that they become the new default. Without this, organisations slide back. The ethical AI initiative gets deprioritised when a new urgency arrives. The training was a one-time event, not a sustained programme. The framework sits in a document, consulted occasionally, ignored routinely.

Refreezing ethical AI transformation means:

Institutionalising new practices: ethical AI checkpoints become standard procedure, not optional. Impact assessments are required before deployment, not recommended. Diversity of perspective in AI teams is a hiring criterion, not a nice-to-have.

Aligning incentives: if people are rewarded purely for speed and scale, ethical AI will always lose to commercial pressure. Refreezing means changing what is measured, what is rewarded, and what is recognised.

Updating culture: culture is what happens when no one is watching. A refrozen ethical AI culture is one where people raise concerns without fear, where shortcuts are questioned, where “we need to think about the impact” is a normal thing to say in a meeting.

Sustaining leadership commitment: ethical AI cannot be a middle-management initiative. It needs visible, sustained commitment from the top. Leaders who model ethical reasoning: who ask hard questions, who slow down when something feels wrong, who protect people who raise concerns: are the most powerful force for refreezing.

Continuous review: AI technology changes rapidly. A framework adequate for today’s AI landscape may be insufficient in two years. Refreezing is not permanent rigidity. It is building the habit of continuous ethical reflection into the organisation’s DNA.

Lewin’s model reminds us that the goal of change is not change for its own sake. It is a new, better equilibrium. For ethical AI, that equilibrium is one where responsible practice is not an extraordinary effort : it is simply how things are done.

Reflection question: What would “refrozen” ethical AI practice look like in your organisation? How would you know it had become the new normal?