All clustering algorithms fail when applied to purely random data with no structure. What specific failure does K-means exhibit in this case?