CipherMind:The Longest Codebook in the World

In recent years, the widespread application of large language models has inspired us to consider using inference for communication encryption. We therefore propose CipherMind, which utilizes intermediate results from deterministic fine-tuning of large model inferences as transmission content. The semantic parameters of large models exhibit characteristics like opaque underlying implementations and weak interpretability, thus enabling their use as an encryption method for data transmission. This communication paradigm can be applied in scenarios like intra-gateway transmission, and theoretically, it can be implemented using any large model as its foundation.

@article{nie2025_2506.15117,
  title={ CipherMind:The Longest Codebook in the World },
  author={ Ming Nie and Zhixiong Yang and Bingsheng Wei },
  journal={arXiv preprint arXiv:2506.15117},
  year={ 2025 }
}