A row security policy is a database rule that limits which table rows a user or query can see or change. Instead of relying only on the application to add the right filter to each query, the database itself applies the rule, so that a query for a table returns only the rows the current user is allowed to see.
It matters in multi-tenant systems because it can act as a second line of defense for keeping each customer's data separate. If the application forgets a filter, the database still stops rows of other tenants from being returned.
It is not a complete answer. Accounts with special privileges, such as superusers or table owners, may bypass the policies unless the database is configured otherwise, and a mistake in a policy can leave data exposed. The rules can also affect performance, and the features and names differ between database systems.
To use them well, enable them on every table that holds tenant data, connect the application with an account that is subject to the policies, test that one tenant cannot read another's rows and review the policies whenever the schema changes.