JPA provides cascades which can be used to automatically persist/update/remove other entities a particular entity depends on. Let's say I have three entities: Job, BankAccount and Person (which has a OneToMany relationship with the first two entities). In the absence of cascade events, I would write:
person.addJob(job);
person.addBankAccount(bankAccount);
em.persist(job);
em.persist(bankAccount);
em.persist(person);
Cascaded events (with the type of CascadeType.PERSIST for instance) make me able to omit first two persist() calls and ensure that a single em.persist(person); call will also persist dependent entities.
Now, this makes the code which acts on entities tightly coupled with their internals. The way I manage entities is now tightly bound to the annotations I used inside the entities. Can this be considered as a bad practice because of the additional coupling it introduces?
Not really, it's fairly standard practice to leverage cascading operations.
You should create integration tests that verify any cascades annotated on your domain actually occur.