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.

Untrustworthy users can misuse image generators to synthesize high-quality deepfakes and engage in unethical activities. Watermarking deters misuse by marking generated content with a hidden message, enabling its detection using a secret watermarking key. A core security property of watermarking is robustness, which states that an attacker can only evade detection by substantially degrading image quality. Assessing robustness requires designing an adaptive attack for the specific watermarking algorithm. When evaluating watermarking algorithms and their (adaptive) attacks, it is challenging to determine whether an adaptive attack is optimal, i.e., the best possible attack. We solve this problem by defining an objective function and then approach adaptive attacks as an optimization problem. The core idea of our adaptive attacks is to replicate secret watermarking keys locally by creating surrogate keys that are differentiable and can be used to optimize the attack's parameters. We demonstrate for Stable Diffusion models that such an attacker can break all five surveyed watermarking methods at no visible degradation in image quality. Optimizing our attacks is efficient and requires less than 1 GPU hour to reduce the detection accuracy to 6.3% or less. Our findings emphasize the need for more rigorous robustness testing against adaptive, learnable attackers.

We present RHODE, a novel system that enables privacy-preserving training of and prediction on Recurrent Neural Networks (RNNs) in a cross-silo federated learning setting by relying on multiparty homomorphic encryption. RHODE preserves the confidentiality of the training data, the model, and the prediction data; and it mitigates federated learning attacks that target the gradients under a passive-adversary threat model. We propose a packing scheme, multi-dimensional packing, for a better utilization of Single Instruction, Multiple Data (SIMD) operations under encryption. With multi-dimensional packing, RHODE enables the efficient processing, in parallel, of a batch of samples. To avoid the exploding gradients problem, RHODE provides several clipping approximations for performing gradient clipping under encryption. We experimentally show that the model performance with RHODE remains similar to non-secure solutions both for homogeneous and heterogeneous data distributions among the data holders. Our experimental evaluation shows that RHODE scales linearly with the number of data holders and the number of timesteps, sub-linearly and sub-quadratically with the number of features and the number of hidden units of RNNs, respectively. To the best of our knowledge, RHODE is the first system that provides the building blocks for the training of RNNs and its variants, under encryption in a federated learning setting.