Yet Another Watermark for Large Language Models

2025-09-15

AI Summary

Existing LLM watermarking methods either rely on sampling adjustments or post-processing—degrading semantic quality—or require fine-tuning under white-box assumptions, incurring high computational overhead. This paper proposes the first parameter-level watermarking framework designed specifically for black-box scenarios: it embeds watermarks via lightweight, reversible internal parameter modulation, achieving deep integration with the model’s intrinsic mechanisms. Crucially, it requires no access to model weights or gradients and enables efficient, zero-access watermark extraction. The method preserves textual semantic fidelity while significantly enhancing robustness against common adversarial attacks—including pruning, paraphrasing, and translation. Experiments demonstrate that its watermark detection accuracy substantially surpasses state-of-the-art sampling- and post-processing-based approaches, all without large-scale fine-tuning. Thus, the framework simultaneously ensures strong copyright protection, practical deployability, and high inference efficiency.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Adversarial Attacks & Robustness
• Natural Language Processing: Safety and Robustness

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Existing watermarking methods for large language models (LLMs) mainly embed watermark by adjusting the token sampling prediction or post-processing, lacking intrinsic coupling with LLMs, which may significantly reduce the semantic quality of the generated marked texts. Traditional watermarking methods based on training or fine-tuning may be extendable to LLMs. However, most of them are limited to the white-box scenario, or very time-consuming due to the massive parameters of LLMs. In this paper, we present a new watermarking framework for LLMs, where the watermark is embedded into the LLM by manipulating the internal parameters of the LLM, and can be extracted from the generated text without accessing the LLM. Comparing with related methods, the proposed method entangles the watermark with the intrinsic parameters of the LLM, which better balances the robustness and imperceptibility of the watermark. Moreover, the proposed method enables us to extract the watermark under the black-box scenario, which is computationally efficient for use. Experimental results have also verified the feasibility, superiority and practicality. This work provides a new perspective different from mainstream works, which may shed light on future research.

Problem

• Develops a watermarking framework for LLMs
• Embeds watermarks via internal parameter manipulation
• Enables black-box extraction without model access

Innovation

• Embedding watermark via internal parameter manipulation
• Black-box extraction without LLM access
• Balancing robustness and imperceptibility efficiently