The widespread deployment of large language models (LLMs) in downstream applications has increased the demand for high-quality system prompts, which have become valuable intellectual assets in the commercial prompt marketplace. Recent studies have demonstrated that system prompts are vulnerable to prompt stealing attacks, where adversaries can extract system prompts from LLM applications by crafting adversarial queries, thereby compromising developers' intellectual property and undermining existing business models. However, prior attack methods suffer from critical limitations including gradient dependency and poor scalability, severely restricting their practical applicability. To address these limitations, we present AGFPS, an automated gradient-free framework that leverages evolutionary optimization to systematically steal prompts. Our approach formulates prompt stealing as a discrete optimization problem, where adversarial queries are modeled as individuals in an evolving population. These individuals are optimized through elite retention, selection, adaptive crossover, and mutation operations. To mitigate local optima convergence, we introduce a progressive fitness evaluation strategy based on adaptive sequence fragmentation that exploits LLMs' autoregressive properties. Comprehensive evaluations across multiple benchmark datasets and mainstream LLMs demonstrates that AGFPS achieves a 95.2% exact system prompt stealing success rate, significantly outperforming manual baselines and surpassing gradient-based methods in 80.6% of scenarios. The generated adversarial queries exhibit remarkable transferability across heterogeneous models and diverse datasets, while maintaining robustness against various defense mechanisms. Our work exposes critical vulnerabilities in current LLM deployment practices and underscores the urgent need for enhanced security measures in LLM applications.

Homomorphic encryption (HE) schemes, such as fully homomorphic encryption (FHE), support a number of useful computations on ciphertext in a broad range of applications, such as e-voting, private information retrieval, cloud security, and privacy protection. While FHE schemes do not require any interaction during computation, the key limitations are large ciphertext expansion and inefficiency. Thus, to overcome these limitations, we develop a novel cryptographic tool, MAS-Encryption (MASE), to support real-value input and secure computation on the multiply-add structure. The multiply-add structures exist in many important protocols, such as classifiers and outsourced protocols, and we will explain how MASE can be used to protect the privacy of these protocols, using two case study examples. Specifically, the first case study example is the privacy-preserving Naive Bayes classifier that can achieve minimal Bayes risk, and the other example is the privacy-preserving support vector machine. We prove that the constructed classifiers are secure and evaluate their performance using real-world datasets. Experiments show that our proposed MASE scheme and MASE based classifiers are efficient, in the sense that we achieve an optimal tradeoff between computation efficiency and communication interactions. Thus, we avoid the inefficiency of FHE based paradigm.

With the advent of cloud computing, data privacy has become one of critical security issues and attracted much attention as more and more mobile devices are relying on the services in cloud. To protect data privacy, users usually encrypt their sensitive data before uploading to cloud servers, which renders the data utilization to be difficult. The ciphertext retrieval is able to realize utilization over encrypted data and searchable public key encryption is an effective way in the construction of encrypted data retrieval. However, the previous related works have not paid much attention to the design of ciphertext retrieval schemes that are secure against inside keyword-guessing attacks (KGAs). In this paper, we first construct a new architecture to resist inside KGAs. Moreover we present an efficient ciphertext retrieval instance with a designated tester (dCRKS) based on the architecture. This instance is secure under the inside KGAs. Finally, security analysis and efficiency comparison show that the proposal is effective for the retrieval of encrypted data in cloud computing.

Abstract An on‐line/off‐line threshold signature (𝒪𝒯𝒮) scheme is a distributed cryptosystem in which a group of players jointly generate a signature for a message and use the on‐line/off‐line technique to improve the efficiency of signing. An 𝒪𝒯𝒮 scheme can be applied to large‐scaled distributed data storage systems and can highly improve the efficiency of writing files. There are two approaches to construct an ordinary threshold signature scheme: the direct approach and the simulation approach. Owing to its simplicity, people tend to use the simulation approach, in which the security of a threshold signature scheme is reduced to the security of its underlying (and simpler) signature scheme. The security proof in this approach is based on a theorem that guarantees the validity of the security reduction—we call this theorem the simulation theorem. However, the simulation theorem (and thus the simulation approach) for an ordinary threshold signature scheme cannot be applied to the on‐line/off‐line cases, because partial signature exposure problems might occur in these cases. This paper presents a simulation theorem for the on‐line/off‐line cases, where the security of an 𝒪𝒯𝒮 scheme is reduced to the security of a so‐called divisible on‐line/off‐line signature scheme. This provides a theoretical basis for constructing an 𝒪𝒯𝒮 scheme through the simulation approach. Furthermore, through this approach, we present a concrete 𝒪𝒯𝒮 scheme, which is efficient and its security proof is simple. Copyright © 2009 John Wiley & Sons, Ltd.

Online/offline signatures are used in a particular scenario where the signer must respond quickly once the message to be signed is presented. In this paper, we present a general method to efficiently convert a trapdoor hash family into an online/offline signature scheme without resorting to any additional signature scheme. We prove that the new scheme is secure in the randomoraclemodel if the underlying trapdoor hash family is collision resistant. Compared to Shamir and Tauman's paradigm, there is an almost 50% reduction in overall computational cost by using the new scheme.