What is the similarity attack against l-diversity?

Question

Accepted Answer

Even with L distinct sensitive values, if they are semantically similar, the attacker learns the sensitive category.
l-diversity treats all values as equally different, but values can be semantically close. Imagine a class whose sensitive values are Gastritis, Gastric Ulcer, Stomach Cancer — three genuinely distinct values, so l-diversity (l=3) is satisfied. Yet an attacker who knows "Dries is under 30" and finds him in that class learns he has a stomach-related disease — a sensitive category — regardless of which exact one.
The flaw: l-diversity protects against guessing the exact value, not against guessing the meaning. This is exactly the gap t-closeness addresses.
Tip: "Diverse" values that all point to the same conclusion aren't really diverse. Semantics matter, not just distinct labels.
Go deeper:

t-closeness (Wikipedia) — the semantic-similarity gap that motivates t-closeness.
 l-diversity (Wikipedia) — l-diversity treats values as equally distinct.