TrojanStego: Your Language Model Can Secretly Be A Steganographic Privacy Leaking Agent

Evidence record

Read the original source ↗

Claim-level citations (2)

• supportsTrojanStego trained a language model to leak secrets inside natural-looking text: For 32-bit secrets, Table 2 reports 87.4% exact recovery with fully fine-tuned Llama-8B when the prompt format was represented in training and the key was unseen. Recovery was 56.1% when both prompt format and key were unseen. Table 6 reports 87-100% exact recovery across four model/training configurations using majority voting over three generations.

  Published v2 paper, Section 6.2 and Table 2, p. 27250; Appendix A.1, Table 6, p. 27256 (https://aclanthology.org/2025.emnlp-main.1386v2.pdf).

• supportsTrojanStego trained a language model to leak secrets inside natural-looking text: TrojanStego fine-tuned an LLM to encode sensitive context information into natural-looking outputs through a vocabulary-partitioning steganographic scheme.

  ACL Anthology abstract