What is the skewness attack, and how does it slip past l-diversity?

Question

Accepted Answer

When a rare value appears more often in a class than in the general population, an attacker learns a probabilistically alarming fact even though l-diversity is technically satisfied.
l-diversity counts distinct values but ignores their base rates. Suppose H1N1 is rare in the population, but 1 of 3 records in an equivalence class has it. The class is "diverse," yet the probability that a person in that class has H1N1 (≈33%) is dramatically higher than in the general public (maybe 0.1%). The attacker has gained meaningful, sensitive information.
So l-diversity can be met on paper while still leaking because it treats the presence of values, not their distribution relative to the baseline.
Tip: Diversity isn't the same as representativeness. Three values where one is a rare disease can be more revealing than they look.
Go deeper:

t-closeness (Wikipedia) — explains the skewness leak l-diversity ignores.
 l-diversity (Wikipedia) — why counting distinct values misses base rates.