Blind quantum computation is a cryptographic primitive that allows a limited-capability client to delegate its complex computation to a remote server without revealing its data and/or computation. This branch of quantum cryptography has been bifurcated into two distinct primitives, quantum homomorphic encryption (concerning the security of only data) and universal blind quantum computation (concerning the security of data and the computing algorithm). These primitives have immense applicability in problems like secure cloud computing, secure quantum variational algorithms, quantum federated learning, and secure multiparty computation. However, no software tools exist for the rapid prototyping of such protocols, hindering the academic interrogation for potential applications. In this paper, we describe the development of the first such library for transpiling circuits written in Qiskit to its blind counterpart, which can then be delegated in a client-server architecture without revealing the client's data and/or computation. The proposed library is designed in modular and reusable component layers, enabling easier scalability to newer BQC primitives and robustness against changes in underlying primitives. We show the implementation of these primitives to a blind variational quantum classifier for the IRIS dataset.
Blind quantum computation is a cryptographic primitive that allows a limited-capability client to delegate its complex computation to a remote server without revealing its data and/or computation. This branch of quantum cryptography has been bifurcated into two distinct primitives, quantum homomorphic encryption (concerning the security of only data) and universal blind quantum computation (concerning the security of data and the computing algorithm). These primitives have immense applicability in problems like secure cloud computing, secure quantum variational algorithms, quantum federated learning, and secure multiparty computation. However, no software tools exist for the rapid prototyping of such protocols, hindering the academic interrogation for potential applications. In this paper, we describe the development of the first such library for transpiling circuits written in Qiskit to its blind counterpart, which can then be delegated in a client-server architecture without revealing the client's data and/or computation. The proposed library is designed in modular and reusable component layers, enabling easier scalability to newer BQC primitives and robustness against changes in underlying primitives. We show the implementation of these primitives to a blind variational quantum classifier for the IRIS dataset.
Blind quantum computing (BQC) is a computational paradigm that allows a client with limited quantum capabilities to delegate quantum computations to a more powerful server while keeping both the algorithm and data hidden. However, in practice, existing BQC protocols face significant challenges when scaling to large-scale computations due to photon losses, low efficiencies, and high overheads associated with fault-tolerant operations, requiring the client to compile both logical operations and error correction primitives. We use a recently demonstrated hybrid light-matter approach [PRL 132, 150604 (2024); Science 388, 509-513 (2025)] to develop an architecture for scalable fault-tolerant blind quantum computation. By combining high-fidelity local gates on the server's matter qubits with delegated blind rotations using photons, we construct loss-tolerant delegated gates that enable efficient algorithm compilation strategies and a scalable approach for fault-tolerant blind logical algorithms. Our approach improves the error-correction threshold and increases the speed and depth of blind logical circuits. Finally, we outline how this architecture can be implemented on state-of-the-art quantum hardware, including neutral atom arrays and solid-state spin defects. These new capabilities open up new opportunities for deep circuit blind quantum computing.
Abstract Quantum computers promise not only to outperform classical machines for certain important tasks, but also to preserve privacy of computation. For example, the blind quantum computing protocol enables secure delegated quantum computation, where a client can protect the privacy of their data and algorithms from a quantum server assigned to run the computation. However, this security comes with the practical limitation that the client and server must communicate after each step of computation. A practical alternative is homomorphic encryption, which does not require any interactions, while providing quantum-enhanced data security for a variety of computations. In this scenario, the server specifies the computation to be performed, and the client provides only the input data, thus enabling secure noninteractive computation. Here, we demonstrate homomorphic-encrypted quantum computing with unitary transformations of individual qubits, as well as multi-qubit quantum walk computations using single-photon states and non-birefringent integrated optics. The client encrypts their input in the photons’ polarization state, while the server performs the computation using the path degree of freedom. Our demonstration using integrated quantum photonics underlines the applicability of homomorphic-encrypted quantum computations, and shows the potential for delegated quantum computing using photons.
Blind Quantum Computation (BQC) enables clients to delegate quantum computations to a quantum server while maintaining the privacy of their data and algorithms, even when the server is untrusted. In this work, we extend BQC frameworks to Quantum Machine Learning (QML) by implementing a network of entangled clients and a quantum server. Specifically, we explore the integration of Variational Quantum Classifiers (VQC) and Quantum Convolutional Neural Networks (QCNNs) within this paradigm. Our proposed model allows clients to perform classical preprocessing and optimization locally while leveraging the quantum server for computationally expensive quantum tasks. The entanglement-based network is managed by a controller that dynamically allocates resources according to the BQC protocol, allowing secure and efficient execution. We present simulation results indicating the feasibility of this approach, including an analysis of network efficiency and resource consumption, alongside the F1 score on QML benchmark datasets.
Blind Quantum Computation lets a limited-capability client delegate its complex computation to a remote server without revealing its data or computation. Several such protocols have been proposed under varied quantum computing models. However, these protocols either rely on highly entangled resource states (in measurement-based models) or are based on non-parametric resource sets (in circuit-based models). These restrictions hinder the practical applicability of such an algorithm in the NISQ era, especially concerning the hybrid quantum-classical infrastructure, which depends on parametric gates. We present a protocol for universal blind quantum computation based on recursive decryption of parametric rotation gates, which does not require a highly entangled state at the server side and substantially reduces the communication rounds required for practical prototyping of secure variational algorithms.
As quantum computing technology matures, an asymmetric computational landscape is emerging where powerful quantum data centers serve users with limited quantum capabilities. The Quantum Internet will enable users to access remote quantum resources, but this delegation raises privacy and trust concerns when the server cannot be assumed to be fully trusted. In this work, we study an implementation-oriented case study in which a Blind Quantum Computing (BQC) protocol is integrated with the encoding stage of a quanvolutional feature extractor for image classification. The goal is to evaluate how an existing BQC-compatible remote state preparation mechanism behaves when used inside a hybrid Quantum Machine Learning workflow. The client encodes classical pixel values as private rotation angles, while fresh random masks hide these angles from an honest-but-curious server at the protocol level. The quantum preprocessing is simulated with finite-shot computational-basis measurements in SquidASM/NetSquid, rather than ideal expectation values, and is evaluated on a small MNIST subset selected for computational feasibility. We analyze the resulting trade-offs among finite-shot sampling, communication and entanglement overhead, depolarizing link and device noise, seed sensitivity, and random-circuit complexity. The study provides practical evidence on the resource and performance constraints that arise when embedding BQC primitives into quantum machine-learning applications, and identifies limitations that must be addressed before deployment-oriented privacy-preserving QML can be considered practical.
Blind quantum computing (BQC) allows a client to delegate quantum computations to a remote server without revealing the input, computation, or output. In addition to being blind, the client can sometimes also verify that the server has performed their instructions correctly, a property known as verifiability. A key part of realizing such verifiable BQC (VBQC) is choosing the design of the client device: many architectures have been proposed, each with different hardware requirements, security properties, and performance characteristics, making it difficult to identify which is most suitable for a given implementation. In this work, we present a comparative analysis of client architectures for VBQC with a matter-qubit server. We restrict our analysis to single-server, single-client protocols with information-theoretic security based on measurement-based quantum computation. We identify three main categories of client: emission-based, measurement-based, and rotation-based, each with multiple variants depending on how the client interacts with the server. We evaluate each across different dimensions: we compare guarantees of existing corresponding security proofs, we derive equations for the rate at which each client can execute a protocol, we provide an overview of each architecture's error behaviour, and discuss hardware cost and design considerations. Client architectures implementing measurement-based remote state preparation and reflection-based teleportation emerge as strong default candidates, but as the right choice remains context-dependent, we provide a framework for navigating considerations to guide the selection of the most suitable architecture for a given setting.