How would you deliberately break K-Means to test its robustness?

You break K-Means by introducing outliers, non-spherical clusters, feature scaling bias, and overlapping distributions.

To test robustness, you intentionally violate K-Means assumptions. Adding extreme outliers shifts centroids dramatically. Non-spherical shapes expose its geometric bias. Unscaled features dominate distance. Overlapping distributions expose hard assignment weaknesses. These stress tests reveal failure boundaries.