JNI: Protect field from dangerous reflective access

I'm designing a Java library that uses a piece of native code exposed via the JNI. The library provides a class that is mapped to a complex native struct that can't easily be serialized. It's not decided yet whether I should store this struct in a ByteBuffer field, or have an int field referring to an index in an array of initialized structs.

In both situations, there's a potential issue where user code could try to modify the ByteBuffer/int, and cause undefined behavior (illegal struct bit representation, out-of-bounds array access, ...). Of course, it isn't in the user's interest to mess with private fields through reflection, but I see Java as a safe language (in terms of UBs or segfaults) and the mere possibility of these kind of issues being rendered possible by my fault annoys me.

Does Java have a way to enforce actual encapsulation, so the fields would actually be completely private? How do other native libraries deal with this, if they even care in the first place?

Thanks for any help.

At the very end, only final primitives cannot be modified through reflection. But to be honest, I have no idea how I could use Java code to modify a data structure that is hold completely outside the JVM and managed only by JNI code (ok, the Java code can call the API provided by the JNI code to modify that data structure, but that's not meant here).

So based on your description, you should be save with the integer used as an index into the 'external' structs.