SQL (that is executed by a client) is audited on the server to which the client connects, and not on servers to which any changes (that result from the SQL) are replicated.

Auditing all database changes is still possible since the changes can be tracked by enabling the AuditInfo feature on all replica servers as well as the master server. However, the complexity that is involved in merging the information into a single audit log increases with number of database nodes in the network.