Garfield-Naruto Encoder V0
Project description:
Our tool allows people to hide messages inside LLM-generated text so that only approved recipients can read or detect them.
We received feedback from several people working in organizing that even the presence or use of encryption tools was being treated as a red flag by government enforcement, and that organizing posts were being taken down from public forums. This breaks the assumption in most cryptography that ciphertext detectability is out of scope for the protocol, so new primitives are needed. Our project is designed to enable community organizing and activism in public forums and scanned messages while remaining totally innocuous to outside observers.
Our solution to this problem was to develop an LLM steganography algorithm that can be decoded by anyone with a secret key. The system uses a model to generate text that blends in with the overwhelming amount of LLM-generated text online today.
We hope this can be used to enable free communication in censored channels and private coordination via large public platforms, and to work against surveillance architecture.
The project is currently looking for funding or contributors to develop a version of the encoder using diffusion models to improve the security and efficiency of the system. Please reach out.
Artifacts:
Contributors:
Violet, Harrison B