We investigate the radioactivity of text generated by large language models (LLM), i.e. whether it is possible to detect that such synthetic input was used to train a subsequent LLM. Current methods like membership inference or active IP protection either work only in settings where the suspected text is known or do not provide reliable statistical guarantees. We discover that, on the contrary, it is possible to reliably determine if a language model was trained on synthetic data if that data is output by a watermarked LLM. Our new methods, specialized for radioactivity, detects with a provable confidence weak residuals of the watermark signal in the fine-tuned LLM. We link the radioactivity contamination level to the following properties: the watermark robustness, its proportion in the training set, and the fine-tuning process. For instance, if the suspect model is open-weight, we demonstrate that training on watermarked instructions can be detected with high confidence ($p$-value $< 10^{-5}$) even when as little as $5\%$ of training text is watermarked.
Invisible Entropy: Towards Safe and Efficient Low-Entropy LLM Watermarking
Mohamed bin Zayed University of Artificial Intelligence（莫扎德·本·扎耶德人工智能大学）；Tsinghua Shenzhen International Graduate School（清华大学深圳国际研究生院）；Tsinghua University（清华大学）；Fudan University（复旦大学）；Shanghai Artificial Intelligence Laboratory（上海人工智能实验室）；University of Notre Dame（Notre Dame 大学）
Can Watermarked LLMs be Identified by Users via Crafted Prompts?
Tsinghua University（清华大学）；Beijing University of Posts and Telecommunications（北京邮电大学）；The Chinese University of Hongkong（香港中文大学）；University of Illinois at Chicago（伊利诺伊大学芝加哥分校）；Hongkong University of Science and Technology (Guangzhou)（香港科技大学（广州））
Leave No TRACE: Black-box Detection of Copyrighted Dataset Usage in Large Language Models via Watermarking
Antiquus S. Hippocampus, Natalia Cerebro & Amelie P. Amygdale Department of Computer Science Cranberry-Lemon University Pittsburgh, PA 15213, USA（计算机科学系，Cranberry-Lemon大学）；Ji Q. Ren & Yevgeny LeNet Department of Computational Neuroscience University of the Witwatersrand Joburg, South Africa（计算神经科学系，沃特瓦特斯兰大学）
Publicly-Detectable Watermarking for Language Models
University of California, Berkeley（加州大学伯克利分校）；University of Wisconsin, Madison（威斯康星大学麦迪逊分校）；Meta；Fundamental Artificial Intelligence Research at Meta（元公司基础人工智能研究部）；University of Virginia（弗吉尼亚大学）；New York University Shanghai（上海纽约大学）
Neural Honeytrace: Plug&Play Watermarking Framework against Model Extraction Attacks
School of Cyberspace Security, Beijing University of Posts and Telecommunications（北京邮电大学网络安全学院）；Cyberspace Institute of Advanced Technology, Guangzhou University（广州大学网络空间技术研究院）；Huangpu Research School of Guangzhou University（广州大学黄埔研究学院）
"Training robust watermarking model may hurt authentication!'' Exploring and Mitigating the Identity Leakage in Robust Watermarking
Zhejiang University（浙江大学）；Hangzhou High-Tech Zone (Binjiang) Institute of Blockchain and Data Security（杭州高新技术产业开发区（滨江）区块链与数据安全研究所）；Tongyi Lab, Alibaba Group（阿里巴巴集团通义实验室）；University of Connecticut（康涅狄格大学）
