Zero-knowledge virtual machine (zkVM) is a powerful infrastructure for proving the correctness of a program execution with a succinct proof, attracting significant interest from researchers, developers, and users. It has been widely used in applications such as blockchain rollups, privacy-preserving machine learning, and off-chain computation. As the field grows, a wide range of zkVMs have been proposed. However, they adopt different choices in instruction formats, trace layouts, and proving backends, which results in a highly heterogeneous design landscape and makes it difficult to understand the relations among these systems.To bridge this gap, we provide a comprehensive study of zkVMs that covers both their theoretical foundations and practical implementations. We decompose zkVMs into three layers: (1) the ISA layer, which defines instruction semantics and determines the structure of the execution trace, (2) the VM layer, which captures program execution and organizes constraints through modular circuit components, and (3) the proving layer, which converts execution traces into algebraic constraints and generates the final proofs. This decomposition allows us to isolate the role of each layer while also examining how they interact in real systems. To give readers a more direct understanding of how these design choices affect performance, scalability, and usability, we conduct a comprehensive experimental evaluation of representative zkVMs following this layered framework. Finally, we conclude the paper by summarizing the main observations from our analysis and outlining several potential directions for zkVM design and implementation.

Zero-knowledge Succinct Non-interactive Argument of Knowledge (zkSNARK) is a powerful cryptographic primitive that enables a prover to convince a verifier that something is true without leaking the private witness. Current zkSNARKs face significant computational costs in generating proofs, which restricts their use in areas like private payments, confidential smart contracts, and anonymous credentials. Private delegation offers a practical solution by outsourcing the heavy computation to powerful external workers without leaking any private information. In this work, we propose HyperSiniel, an efficient private delegation framework for general zkSNARKs that achieves a new feature called guaranteed output delivery (GOD). HyperSiniel is designed to be compatible with any universal zkSNARKs constructed from a polynomial interactive oracle proof (PIOP) and a polynomial commitment scheme (PCS). It enables a computationally limited delegator to outsource proof generation to several workers in a fully non-interactive and privacy-preserving manner. Compared to the most state-of-the-art frameworks (e.g., Siniel [NDSS'25]), HyperSiniel ensures that the delegator always receives a correct proof, regardless of malicious worker behavior. We implement HyperSiniel and compare the performance with Siniel across varying bandwidths and circuit sizes. Under low-bandwidth conditions (10MBps), HyperSiniel incurs only an additional 25% overhead compared with Siniel, while the total running time of HyperSiniel is almost identical to Siniel under high-bandwidth settings (1000MBps). These results show that the strong robustness guarantee of GOD in HyperSiniel comes almost for free, making it a practical and secure solution for real-world zkSNARK delegation.

Data duplication within large-scale corpora often impedes large language models' (LLMs) performance and privacy. In privacy-concerned federated learning scenarios, conventional deduplication methods typically rely on trusted third parties to perform uniform deletion, risking loss of informative samples while introducing privacy vulnerabilities. To address these gaps, we propose Federated ReWeighting (FedRW), the first privacy-preserving framework, to the best of our knowledge, that performs soft deduplication via sample reweighting instead of deletion in federated LLM training, without assuming a trusted third party. At its core, FedRW proposes a secure, frequency-aware reweighting protocol through secure multi-party computation, coupled with a parallel orchestration strategy to ensure efficiency and scalability. During training, FedRW utilizes an adaptive reweighting mechanism with global sample frequencies to adjust individual loss contributions, effectively improving generalization and robustness. Empirical results demonstrate that FedRW outperforms the state-of-the-art method by achieving up to $28.78\times$ speedup in preprocessing and approximately $11.42$\% improvement in perplexity, while offering enhanced security guarantees. FedRW thus establishes a new paradigm for managing duplication in federated LLM training.

The Internet of Things (IoT) has greatly changed our lives and generated a large amount of data. Cloud storage helps IoT limited-resource IOT devices process the massive data. However, cloud servers are untrusted in most scenarios as they may illegally obtain sensitive data. Although existing symmetric searchable encryption (SSE) schemes can protect the privacy of outsourced data while preserve data availability, Most of them leak access and search patterns to the cloud server to gain better performance. Such leakages will be used to recover private information. Meanwhile, semi-honestly secure searchable encryption cannot prevent attacks done by the malicious server such as returning the false search result. Therefore, it is still a challenge to prevent malicious cloud server misbehavior, and preserve patterns, simultaneously. This paper proposes OpenSE to solve the aforementioned problems. First, this paper constructs FastOPE as a major building block. With the OPE protocol, the verifiable searchable encryption OpenSE can be trivially realized. After that, security proofs show that OpenSE is secure against malicious cloud servers with access and search pattern hidden. Finally, we implement experiments on real datasets to compare OpenSE with some state-of-the-art works in terms of running time of setup phase and search phase as well as storage overhead. The experimental results show that OpenSE outperforms the state-of-the-art works in terms of setup phase and storage overhead. In addition, the theoretic comparison shows that OpenSE outperforms most existing works in terms of security, in which OpenSE enjoys both verifiability and pattern hidden.

Federated learning (FL) allows multiple parties, each holding a dataset, to jointly train a model without leaking any information about their own datasets. In this paper, we focus on vertical FL (VFL). In VFL, each party holds a dataset with the same sample space and different feature spaces. All parties should first agree on the training dataset in the ID alignment phase. However, existing works may leak some information about the training dataset and cause privacy leakage. To address this issue, this paper proposes OpenVFL, a vertical federated learning framework with stronger privacy-preserving. We first propose NCLPSI, a new variant of labeled PSI, in which both parties can invoke this protocol to get the encrypted training dataset without leaking any additional information. After that, both parties train the model over the encrypted training dataset. We also formally analyze the security of OpenVFL. In addition, the experimental results show that OpenVFL achieves the best trade-offs between accuracy, performance, and privacy among the most state-of-the-art works.

Searchable encryption (SE) allows a data owner to outsource encrypted documents to an untrusted cloud server while preserving privacy and achieving secure data sharing. However, most existing SE schemes have a trade-off between security and efficiency. Moreover, these SE schemes leak the server's partial database or search information to perform better. Recent attacks show that such leakages can be used to recover the content of queried keywords or partial database information. To solve this problem and ensure efficiency, this paper proposes labeled searchable encryption (LSE), an efficient searchable encryption scheme based on the labeled private set intersection. We also give formal proofs to prove the security of the proposed labeled PSI protocol and searchable encryption scheme. Finally, we do experiments to compare the performance with some state-of-the-art works, and the experimental results show that the LSE outperforms in terms of total size and generation time of the encrypted database as well as the total search time at client side.

Multiparty private set intersection (PSI) allows several parties, each holding a set of elements, to jointly compute the intersection without leaking any additional information. With the development of cloud computing, delegating the computation to an untrsuted cloud server is becoming a major problem, where the untrusted cloud server may try to get some sensitive information from clients' private information. However, it is complex to build an efficient and reliable scheme to protect user privacy. In order to overcome this problem, we propose DMPSI, an efficient delegated PSI (with cardinality) protocol in a multiparty setting. DMPSI avoids using heavy cryptographic primitives (mainly rely on symmetric-key encryption) to achieve better performance. In addition, both PSI and PSI with the cardinality of DMPSI are secure against semi-honest adversaries and allow any number of colluding clients (at least one honest client). We do experiments to compare the proposed DMPSI with some state-of-the-art works to evaluate overall performance. In addition, we also compare the proposed Oks-PRF with some state-of-the-art multi-point OPRF to highlight our efficiency. The experimental results show that proposed both Oks-PRF and DMPSI(-CA) has better performance and is scalable in the number of clients and the set size.

Private Set Intersection (PSI) is one of the most important functions in secure multiparty computation (MPC). PSI protocols have been a practical cryptographic primitive and there are many privacy-preserving applications based on PSI protocols such as computing conversion of advertising and distributed computation. Private Set Intersection Cardinality (PSI-CA) is a useful variant of PSI protocol. PSI and PSI-CA allow several parties, each holding a private set, to jointly compute the intersection and cardinality, respectively without leaking any additional information. Nowadays, most PSI protocols mainly focus on two-party settings, while in multiparty settings, parties are able to share more valuable information and thus more desirable. On the other hand, with the advent of cloud computing, delegating computation to an untrusted server becomes an interesting problem. However, most existing delegated PSI protocols are unable to efficiently scale to multiple clients. In order to solve these problems, this paper proposes MDPPC, an efficient PSI protocol which supports scalable multiparty delegated PSI and PSI-CA operations. Security analysis shows that MDPPC is secure against semi-honest adversaries and it allows any number of colluding clients. For 15 parties with set size of 220on server side and 216on clients side, MDPPC costs only 81 seconds in PSI and 80 seconds in PSI-CA, respectively. The experimental results show that MDPPC has high scalability.