Optimizing Adaptive Attacks against Watermarks for Language Models

Abdulrahman Diaa, Toluwani Aremu, Nils Lukas
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:13546-13569, 2025.

Abstract

Large Language Models (LLMs) can be misused to spread unwanted content at scale. Content watermarking deters misuse by hiding messages in content, enabling its detection using a secret watermarking key. Robustness is a core security property, stating that evading detection requires (significant) degradation of the content’s quality. Many LLM watermarking methods have been proposed, but robustness is tested only against non-adaptive attackers who lack knowledge of the watermarking method and can find only suboptimal attacks. We formulate watermark robustness as an objective function and use preference-based optimization to tune adaptive attacks against the specific watermarking method. Our evaluation shows that (i) adaptive attacks evade detection against all surveyed watermarks, (ii) training against any watermark succeeds in evading unseen watermarks, and (iii) optimization-based attacks are cost-effective. Our findings underscore the need to test robustness against adaptively tuned attacks. We release our adaptively tuned paraphrasers at https://github.com/nilslukas/ada-wm-evasion.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-diaa25a, title = {Optimizing Adaptive Attacks against Watermarks for Language Models}, author = {Diaa, Abdulrahman and Aremu, Toluwani and Lukas, Nils}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {13546--13569}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/diaa25a/diaa25a.pdf}, url = {https://proceedings.mlr.press/v267/diaa25a.html}, abstract = {Large Language Models (LLMs) can be misused to spread unwanted content at scale. Content watermarking deters misuse by hiding messages in content, enabling its detection using a secret watermarking key. Robustness is a core security property, stating that evading detection requires (significant) degradation of the content’s quality. Many LLM watermarking methods have been proposed, but robustness is tested only against non-adaptive attackers who lack knowledge of the watermarking method and can find only suboptimal attacks. We formulate watermark robustness as an objective function and use preference-based optimization to tune adaptive attacks against the specific watermarking method. Our evaluation shows that (i) adaptive attacks evade detection against all surveyed watermarks, (ii) training against any watermark succeeds in evading unseen watermarks, and (iii) optimization-based attacks are cost-effective. Our findings underscore the need to test robustness against adaptively tuned attacks. We release our adaptively tuned paraphrasers at https://github.com/nilslukas/ada-wm-evasion.} }
Endnote
%0 Conference Paper %T Optimizing Adaptive Attacks against Watermarks for Language Models %A Abdulrahman Diaa %A Toluwani Aremu %A Nils Lukas %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-diaa25a %I PMLR %P 13546--13569 %U https://proceedings.mlr.press/v267/diaa25a.html %V 267 %X Large Language Models (LLMs) can be misused to spread unwanted content at scale. Content watermarking deters misuse by hiding messages in content, enabling its detection using a secret watermarking key. Robustness is a core security property, stating that evading detection requires (significant) degradation of the content’s quality. Many LLM watermarking methods have been proposed, but robustness is tested only against non-adaptive attackers who lack knowledge of the watermarking method and can find only suboptimal attacks. We formulate watermark robustness as an objective function and use preference-based optimization to tune adaptive attacks against the specific watermarking method. Our evaluation shows that (i) adaptive attacks evade detection against all surveyed watermarks, (ii) training against any watermark succeeds in evading unseen watermarks, and (iii) optimization-based attacks are cost-effective. Our findings underscore the need to test robustness against adaptively tuned attacks. We release our adaptively tuned paraphrasers at https://github.com/nilslukas/ada-wm-evasion.
APA
Diaa, A., Aremu, T. & Lukas, N.. (2025). Optimizing Adaptive Attacks against Watermarks for Language Models. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:13546-13569 Available from https://proceedings.mlr.press/v267/diaa25a.html.

Related Material