HiTMS: A High-Throughput Multi-Stream Linguistic Steganography Framework
This paper introduces HiTMS, a method for concealing secrets in large language models across multiple responses in successive rounds, improving throughput and reducing steganalyzer detection.
Proposed a new method, HiTMS, for concealing secrets in large language models by distributing the secret across multiple responses in successive rounds.
Keywords
Before reading this…
Applications
- →Data security
- →Communication security
To understand this paper, make sure you know these concepts first:
- Understanding of large language modelsfind papers →
- Concept of steganographyfind papers →
Abstract
More Like ThisGenerative linguistic steganography conceals secret bits within the sampling randomness of large language models. Existing schemes are single-stream, conveying an entire secret through a single response to a single prompt. This convention incurs two limitations: it provides no protocol-level support for batched multi-stream inference, and naive co-batching does not conceal slot occupancy or payload completion. We propose HiTMS, which distributes a secret across multiple responses produced jointly over successive rounds of interaction. Each round embeds and extracts several streams within a single batched call, thereby amortizing the cost of model invocation and substantially improving throughput. To ensure recoverability, HiTMS wraps each response in a self-describing frame and employs a key-derived schedule that binds streams to slots and fills unused slots with decoys, guaranteeing exact recovery while concealing the number of active streams. The framework is agnostic to both the language model and the steganographic coder. Across eight dataset-model-coder settings, eight-stream HiTMS achieves up to 4.3 times higher embedding and extraction speeds than single-stream baselines, while reducing the steganalyzer AUROC from 0.681 to 0.601 on average. Additional experiments with 4 to 64 streams demonstrate sustained throughput gains as concurrency increases. GitHub repository for this work is https://github.com/ryehr/HiTMS_steganography.