Two different ways to bound what an aggregate reveals
K-anonymity requires that every record in a released dataset be indistinguishable, on identifying fields, from at least a set number of other records — generalizing or suppressing values until every group is large enough that no individual within it can be pinpointed. It's a concrete, measurable standard, but it has a known limit: if most of a group happens to share the same value on some other, sensitive field, that value can still be inferred for the whole group even without identifying which specific person it applies to.
Differential privacy takes a different approach, adding carefully calibrated statistical noise to query results so that no single individual's presence in the data meaningfully changes the answer — a guarantee that holds even against someone running many different queries and comparing results to triangulate an individual. The tradeoff is direct: more noise means a stronger guarantee and a less precise answer, and getting that balance right for a specific use case is a deliberate design decision, not a default setting appropriate everywhere.
Why aggregation alone isn't a safe default assumption
Neither technique is automatic — applying no disclosure-limiting technique at all and simply assuming that 'it's aggregated, so it's fine' is exactly the assumption that leads to individuals being re-identified from data nobody thought was individually attributable. The safety has to be actually engineered into how the aggregate is constructed, not inferred from the fact that individual records aren't directly visible in the output.