Advances in generative models have made it possible for AI-generated text, code, and images to mirror human-generated content in many applications. W atermark-ing, a technique that aims to embed information in the output of a model to verify its source, is useful for mitigating the misuse of such AI-generated content. However, we show that common design choices in LLM watermarking schemes make the resulting systems surprisingly susceptible to attack--leading to fundamental trade-offs in robustness, utility, and usability. To navigate these trade-offs, we rigorously study a set of simple yet effective attacks on common watermarking systems, and propose guidelines and defenses for LLM watermarking in practice.

Advances in generative models have made it possible for AI-generated text, code, and images to mirror human-generated content in many applications. W atermark-ing, a technique that aims to embed information in the output of a model to verify its source, is useful for mitigating the misuse of such AI-generated content. However, we show that common design choices in LLM watermarking schemes make the resulting systems surprisingly susceptible to attack--leading to fundamental trade-offs in robustness, utility, and usability. To navigate these trade-offs, we rigorously study a set of simple yet effective attacks on common watermarking systems, and propose guidelines and defenses for LLM watermarking in practice.

In recent years, LLM watermarking has emerged as an attractive safeguard against AI-generated content, with promising applications in many real-world domains. However, there are growing concerns that the current LLM watermarking schemes are vulnerable to expert adversaries wishing to reverse-engineer the watermarking mechanisms. Prior work in "breaking" or "stealing" LLM watermarks mainly focuses on the distribution-modifying algorithm of Kirchenbauer et al. (2023), which perturbs the logit vector before sampling. In this work, we focus on reverse-engineering the other prominent LLM watermarking scheme, distortion-free watermarking (Kuditipudi et al. 2024), which preserves the underlying token distribution by using a hidden watermarking key sequence. We demonstrate that, even under a more sophisticated watermarking scheme, it is possible to "compromise" the LLM and carry out a "spoofing" attack. Specifically, we propose a mixed integer linear programming framework that accurately estimates the secret key used for watermarking using only a few samples of the watermarked dataset. Our initial findings challenge the current theoretical claims on the robustness and usability of existing LLM watermarking techniques.

As artificial intelligence surpasses human capabilities in text generation, the necessity to authenticate the origins of AI-generated content has become paramount. Unbiased watermarks offer a powerful solution by embedding statistical signals into language model-generated text without distorting the quality. In this paper, we introduce MCmark, a family of unbiased, Multi-Channel-based watermarks. MCmark works by partitioning the model's vocabulary into segments and promoting token probabilities within a selected segment based on a watermark key. We demonstrate that MCmark not only preserves the original distribution of the language model but also offers significant improvements in detectability and robustness over existing unbiased watermarks. Our experiments with widely-used language models demonstrate an improvement in detectability of over 10% using MCmark, compared to existing state-of-the-art unbiased watermarks. This advancement underscores MCmark's potential in enhancing the practical application of watermarking in AI-generated texts.

Text watermarking for Large Language Models (LLMs) has made significant progress in detecting LLM outputs and preventing misuse. Current watermarking techniques offer high detectability, minimal impact on text quality, and robustness to text editing. However, current researches lack investigation into the imperceptibility of watermarking techniques in LLM services. This is crucial as LLM providers may not want to disclose the presence of watermarks in real-world scenarios, as it could reduce user willingness to use the service and make watermarks more vulnerable to attacks. This work investigates the imperceptibility of watermarked LLMs. We design the first unified identification method called Water-Probe that identifies all kinds of watermarking in LLMs through well-designed prompts. Our key motivation is that current watermarked LLMs expose consistent biases under the same watermark key, resulting in similar differences across prompts under different watermark keys. Experiments show that almost all mainstream watermarking algorithms are easily identified with our well-designed prompts, while Water-Probe demonstrates a minimal false positive rate for non-watermarked LLMs. Finally, we propose that the key to enhancing the imperceptibility of watermarked LLMs is to increase the randomness of watermark key selection. Based on this, we introduce the Water-Bag strategy, which significantly improves watermark imperceptibility by merging multiple watermark keys. The rapid advancement of large language models (LLMs) has led to remarkable achievements in tasks such as question answering (Zhuang et al., 2024), programming (Jiang et al., 2024b), and reasoning (Wei et al., 2022), with widespread applications across various scenarios. Recent research indicates that malicious attackers can steal LLMs through model extraction techniques (Yao et al., 2024), and some users may abuse LLMs to generate and spread harmful information (Wei et al., 2024). Text watermarking techniques for LLMs have become an important method to mitigate the above issues by adding detectable features to LLM outputs (Liu et al., 2024b). Recent researches on LLM watermarking have focused on improving watermark detectability (Kirchenbauer et al., 2023a), minimizing impact on generated text (Aaronson & Kirchner, 2022), and enhancing robustness against text modifications (Liu et al., 2024a).

Generative models have enabled easy creation and generation of images of all kinds given a single prompt. However, this has also raised ethical concerns about what is an actual piece of content created by humans or cameras compared to model-generated content like images or videos. Watermarking data generated by modern generative models is a popular method to provide information on the source of the content. The goal is for all generated images to conceal an invisible watermark, allowing for future detection or identification. The Stable Signature finetunes the decoder of Latent Diffusion Models such that a unique watermark is rooted in any image produced by the decoder. In this paper, we present a novel adversarial fine-tuning attack that disrupts the model's ability to embed the intended watermark, exposing a significant vulnerability in existing watermarking methods. To address this, we further propose a tamper-resistant fine-tuning algorithm inspired by methods developed for large language models, tailored to the specific requirements of watermarking in LDMs. Our findings emphasize the importance of anticipating and defending against potential vulnerabilities in generative systems.

Language model (LM) watermarking techniques inject a statistical signal into LM-generated content by substituting the random sampling process with pseudo-random sampling, using watermark keys as the random seed. Among these statistical watermarking approaches, distortion-free watermarks are particularly crucial because they embed watermarks into LM-generated content without compromising generation quality. However, one notable limitation of pseudo-random sampling compared to true-random sampling is that, under the same watermark keys (i.e., key collision), the results of pseudo-random sampling exhibit correlations. This limitation could potentially undermine the distortion-free property. Our studies reveal that key collisions are inevitable due to the limited availability of watermark keys, and existing distortion-free watermarks exhibit a significant distribution bias toward the original LM distribution in the presence of key collisions. Moreover, achieving a perfect distortion-free watermark is impossible as no statistical signal can be embedded under key collisions. To reduce the distribution bias caused by key collisions, we introduce a new family of distortion-free watermarks--beta-watermark. Experimental results support that the beta-watermark can effectively reduce the distribution bias under key collisions.

Large language models (LLMs) excellently generate human-like text, but also raise concerns about misuse in fake news and academic dishonesty. Decoding-based watermark, particularly the GumbelMax-trick-based watermark(GM watermark), is a standout solution for safeguarding machine-generated texts due to its notable detectability. However, GM watermark encounters a major challenge with generation diversity, always yielding identical outputs for the same prompt, negatively impacting generation diversity and user experience. To overcome this limitation, we propose a new type of GM watermark, the Logits-Addition watermark, and its three variants, specifically designed to enhance diversity. Among these, the GumbelSoft watermark (a softmax variant of the Logits-Addition watermark) demonstrates superior performance in high diversity settings, with its AUROC score outperforming those of the two alternative variants by 0.1 to 0.3 and surpassing other decoding-based watermarking methods by a minimum of 0.1.

Advances in generative models have made it possible for AI-generated text, code, and images to mirror human-generated content in many applications. Watermarking, a technique that aims to embed information in the output of a model to verify its source, is useful for mitigating the misuse of such AI-generated content. However, we show that common design choices in LLM watermarking schemes make the resulting systems surprisingly susceptible to attack -- leading to fundamental trade-offs in robustness, utility, and usability. To navigate these trade-offs, we rigorously study a set of simple yet effective attacks on common watermarking systems, and propose guidelines and defenses for LLM watermarking in practice.

Engineering a top-notch deep learning model is an expensive procedure that involves collecting data, hiring human resources with expertise in machine learning, and providing high computational resources. For that reason, deep learning models are considered as valuable Intellectual Properties (IPs) of the model vendors. To ensure reliable commercialization of deep learning models, it is crucial to develop techniques to protect model vendors against IP infringements. One of such techniques that recently has shown great promise is digital watermarking. However, current watermarking approaches can embed very limited amount of information and are vulnerable against watermark removal attacks. In this paper, we present GradSigns, a novel watermarking framework for deep neural networks (DNNs). GradSigns embeds the owner's signature into the gradient of the cross-entropy cost function with respect to inputs to the model. Our approach has a negligible impact on the performance of the protected model and it allows model vendors to remotely verify the watermark through prediction APIs. We evaluate GradSigns on DNNs trained for different image classification tasks using CIFAR-10, SVHN, and YTF datasets. Experimental results show that GradSigns is robust against all known counter-watermark attacks and can embed a large amount of information into DNNs.