https://arxiv.org/abs/2505.18332 An Attack to Break Permutation-Based Private Third-Party Inference Schemes for LLMsRecent advances in Large Language Models (LLMs) have led to the widespread adoption of third-party inference services, raising critical privacy concerns. Existing methods of performing private third-party inference, such as Secure Multiparty Computation (Sarxiv.orghidden state를 per..