Provable Secure Steganography Based on Adaptive Dynamic Sampling

The security of private communication is increasingly at risk due to widespread surveillance. Steganography, a technique for embedding secret messages within innocuous carriers, enables covert communication over monitored channels. Provably Secure Steganography (PSS) is state of the art for making stego carriers indistinguishable from normal ones by ensuring computational indistinguishability between stego and cover distributions. However, current PSS methods often require explicit access to the distribution of generative model for both sender and receiver, limiting their practicality in black box scenarios. In this paper, we propose a provably secure steganography scheme that does not require access to explicit model distributions for both sender and receiver. Our method incorporates a dynamic sampling strategy, enabling generative models to embed secret messages within multiple sampling choices without disrupting the normal generation process of the model. Extensive evaluations of three real world datasets and three LLMs demonstrate that our blackbox method is comparable with existing white-box steganography methods in terms of efficiency and capacity while eliminating the degradation of steganography in model generated outputs.

Learnable Linguistic Watermarks for Tracing Model Extraction Attacks on Large Language Models

In the rapidly evolving domain of artificial intelligence, safeguarding the intellectual property of Large Language Models (LLMs) is increasingly crucial. Current watermarking techniques against model extraction attacks, which rely on signal insertion in model logits or post-processing of generated text, remain largely heuristic. We propose a novel method for embedding learnable linguistic watermarks in LLMs, aimed at tracing and preventing model extraction attacks. Our approach subtly modifies the LLM's output distribution by introducing controlled noise into token frequency distributions, embedding an statistically identifiable controllable watermark.We leverage statistical hypothesis testing and information theory, particularly focusing on Kullback-Leibler Divergence, to differentiate between original and modified distributions effectively. Our watermarking method strikes a delicate well balance between robustness and output quality, maintaining low false positive/negative rates and preserving the LLM's original performance.