Secure aggregation enhances the security of federated learning systems by:

Secure aggregation cryptographically aggregates client updates so that the server learns only the sum/aggregate, not any single client’s update. Properly implemented, the server cannot recover individual contributions—even if compromised—thereby preserving client confidentiality. Option C (encryption in transit) is insufficient because decryption at the server reveals updates; Option A is procedural, not cryptographic; Option B (differential privacy) is a separate technique and not the defining property of secure aggregation.