aegistudio

Hmm... LLMs / MLLMs might be truly a unified input / output interface of a would-be AGI, I think.

You are right. I don't come with cryptography background and haven't consider the attack model carefully. I will warn the users about the current status and strive to design it to confront potential attacks.

Sure. Thanks for the reference.

May I repeat your statements in my words, so that we can have understanding in common and communicate efficiently?

1. Elaborated and formalized description about encryption method is required.

2. Processes need to be authenticated before they could access ciphertext / plaintext directory, so that they can't recover the key stream.

3. On encrypting the file names, the threshold of directory tree size for the birthday attack to become innegligible, should also be elaborated.