If you load malicious code into your address space and execute it, then it can always do malicious things to your data. If you load malicious code into a separate process and execute it, then it can almost certainly do malicious things to your data, unless you put it into a locked-down user context and trust your OS and environment not to have any local privilege escalations (lol). The only real way to load untrusted native code is to put it in an OS-level container and communicate via IPC, or better yet, put it in a VM and communicate via a virtual network.

The measures I suggested before were all in the context of buggy users that can't resist the urge to keep references to the library's resources lying around all over the place. But untrusted code can never be made safe with anything short of a strong sandbox.