The field of linguistic steganography and watermarking is moving towards developing more secure and efficient methods for embedding and detecting hidden information in text. Recent research has focused on addressing the challenges of tokenization ambiguity, improving embedding capacity, and enhancing the robustness of watermarking algorithms. Notably, innovative approaches such as context-aware thresholding and order-agnostic watermarking have been proposed to improve the performance of watermarking algorithms in various scenarios.

Some papers are particularly noteworthy for their innovative contributions. For example, one paper proposes a novel disambiguation algorithm that overcomes the capacity limitation of existing methods while retaining provable security guarantees. Another paper introduces a context-aware thresholding framework that dynamically adjusts watermarking intensity based on real-time semantic context, improving text quality in cross-tasks without sacrificing detection accuracy. A third paper presents the first watermarking framework designed specifically for diffusion large language models, achieving high detection rates while maintaining text quality. Additionally, a paper on agent-driven strategy evolution in LLM-based text steganography demonstrates superior performance with gains in perplexity and anti-steganalysis performance over state-of-the-art methods.