Shared quantum-cloud runtimes expose service outcomes shaped by compilation, backend conditions, and scheduling. We study whether workload-dependent information remains in tenant-visible outputs after scheduler mediation, while separating such evidence from provider-side boundary traces and internal diagnostics. Experiments use a provider-inspired prototype with synthetic IBM-style backend abstractions rather than IBM’s proprietary scheduler. Across 16 independent workload seeds, leave-one-seed-out evaluation using provider-reported waiting time yields a mean accuracy of 0.6010 and a classification-based signed advantage of 0.1010 over the majority-class baseline. Monotone padding and delayed release reduce the estimated advantage to 0.0324 and 0.0418, with mean added delays of 0.185 and 0.131 in normalized simulator-time units. Controlled workload pairs and compilation- and backend-level diagnostics show configuration-dependent, nonmonotonic variation across internal observability channels, but do not attribute the external signal to a uniquely quantum mechanism. A six-run experiment on one public backend provides only weak and variable in-sample timing separability. These results demonstrate preliminary scheduler-boundary separability in the controlled prototype, but do not establish a practical public-cloud attack.
Shared quantum-cloud runtimes expose service outcomes shaped by compilation, backend conditions, and scheduling. We study whether workload-dependent information remains in tenant-visible outputs after scheduler mediation, while separating such evidence from provider-side boundary traces and internal diagnostics. Experiments use a provider-inspired prototype with synthetic IBM-style backend abstractions rather than IBM’s proprietary scheduler. Across 16 independent workload seeds, leave-one-seed-out evaluation using provider-reported waiting time yields a mean accuracy of 0.6010 and a classification-based signed advantage of 0.1010 over the majority-class baseline. Monotone padding and delayed release reduce the estimated advantage to 0.0324 and 0.0418, with mean added delays of 0.185 and 0.131 in normalized simulator-time units. Controlled workload pairs and compilation- and backend-level diagnostics show configuration-dependent, nonmonotonic variation across internal observability channels, but do not attribute the external signal to a uniquely quantum mechanism. A six-run experiment on one public backend provides only weak and variable in-sample timing separability. These results demonstrate preliminary scheduler-boundary separability in the controlled prototype, but do not establish a practical public-cloud attack.
Remote quantum computing exposes a confidentiality gap. Standard privacy mechanisms protect quantum states and outputs, but not the scientific structure of a workload. This work reveals that hardware-aware compilation leaves observable signatures, such as routing overhead, circuit depth, and gate composition, that correlate with hidden modelling choices like partial differential equation boundary conditions, discretisation scale, and molecular geometry. The leakage arises from the mismatch between logical topology and fixed hardware connectivity, forcing problem-dependent SWAP insertion. We formalise this threat as Scientific-Intent Indistinguishability and prove that passive security is asymptotically unachievable under routing-optimal compilation. Experiments on a 156-qubit IBM Heron processor achieve near-perfect classification of boundary regimes and molecular geometries, with leakage generalising across solver families via routing-scaling exponents. Conventional gate-padding fails as a defence, causing fidelity drops without reducing adversarial advantage. Our results show that protecting quantum data alone is insufficient; execution-level confidentiality must become a first-class design requirement.
We study a quantum-runtime hypothesis that is narrower than“betterqubits”andbroader than ordinary hardware-aware compilation: distinct subroutines of a quantum program may require different protection profiles and different qubit-array geometries because they carry different kinds of informational fragility. The runtime considered here classifies circuit windows into phase-sensitive, globally entangled, arithmetic, weak-signal, magic-state, and stabilizer-oriented workloads, and then allocates both protection and geometry accordingly. We evaluate this idea in simulation using hardware-informed calibration inputs, structured logical noise, explicit geometry profiles, and paired condition randomness so that ablation modes face the same seeded noise realization. The focused three-regime study uses 96 replicates across three superconducting backends and compares four runtime modes: uniform, protection-only, geometry-only, and protection-plus-geometry, each under static and workload-reactive noise. Geometry is not a universal accelerator. Instead, it behaves as a selective coherence-control variable. In the paired geometry-on-top-of-protection comparison, the strongest gains occur for global parity and weak-signal spectroscopy under mixed workload-reactive noise, with mean useful-output improvements of approximately +0.188 and +0.181, respectively. A stabilizer-specific geometry refinement converts an earlier regression into a helpful-clean result in the same regime.
Abstract Quantum cloud users access remote processors through classical measurement transcripts. The repeated shots that are operationally required for reliable quantum readout can systematically resolve microscopic, quasi-static hardware variations into stable statistical fingerprints. On IBM superconducting processors, we show that a transcript-only adversary can identify the backend with up to 89.3% attack success rate at T=20 repetition rounds under multi-interface aggregation. This exposes a privacy locus mismatch: many channel-centric quantum privacy formulations regulate distinguishability at the level of quantum states or channels, whereas the operational attack in Quantum-Computing-as-a-Service (QCaaS) targets repeated classical transcripts and the latent service provenance they reveal after measurement. We therefore formulate privacy for repeated quantum workloads as a cross-layer provenance-protection problem. We protect the latent service provenance behind a released transcript, namely the service provenance that jointly reflects backend realization and serving context in QCaaS settings where this provenance is not fully disclosed to the observing party. At the measurement interface, we formulate Measurement-Agnostic Quantum Differential Privacy (MA-QDP) as an operational transcript-level privacy criterion for repeated quantum workloads, uniform over physically admissible POVMs and arbitrary classical post-processing. MA-QDP specifies the privacy locus and protected secret appropriate to QCaaS transcript release, and is instantiated via atomic outcome randomization before aggregation. At the orchestration layer, we incorporate context decorrelation to suppress persistent linkage of repeated sessions to a stable service provenance. In controlled simulations, the integrated framework suppresses provenance inference under realistic shot budgets while preserving task-relevant information, yielding a favorable privacy-utility frontier compared to channel-level and classical-level perturbations. Together, our results identify repetition as an intrinsic fingerprint amplifier in QCaaS and motivate a cross-layer provenance-protection architecture for secure, repeated quantum workloads.
Quantum cloud platforms increasingly provide access to heterogeneous execution backends including real QPUs and quantum simulators. Effective backend selection is challenging because runtime and output quality vary with workload structure, backend characteristics, and noise conditions. This paper presents an empirical characterization study of how quantum workload properties shape execution across heterogeneous classical simulator backends, focusing on a CPU/GPU-based statevector simulation method. The workload characteristics examined in this study include qubit count, circuit depth, two-qubit gate density, workload class, and noise sensitivity. The experiments show qubit count and circuit depth are primary factors for simulation cost. Two-qubit gate density and workload class affect backend affinity. For example, CPU/GPU affinity in Grover varies with the depth of circuits. Shallow circuits have lower runtime on CPUs, whereas deep circuits have lower runtime on GPUs. Four backend-selection criteria (BS1-BS4) provide analytical bases to examine the effects of circuit structure, workload class, noise tiers, and measurement-derived profiles. The results show characterization-driven backend selection performs better than relying on coarse-grained workload information alone in heterogeneous quantum simulation environments.
Present-day quantum computing is cloud-based, where a user submits a circuit to a service provider's proprietary backend hardware. While providers may wish to hide implementation details, scheduling choices, or even which physical device was used, noisy finite-shot outputs can carry backend-specific fingerprints: information imprinted in the classical output distribution that can reveal the backend identity. So far, such fingerprints have mostly been studied from a benchmarking perspective, with limited attention to privacy considerations for users and providers. This work develops the first formal framework for backend identifiability and its privacy implications. We introduce a backend-identifiability game and use it to formalise routing anonymity as a security notion for quantum cloud services. We show that backend identifiability is a hypothesis-testing problem and prove that, under passive i.i.d. access to a single backend, routing anonymity decays exponentially at the Chernoff rate. We also establish a utility-anonymity trade-off, imposing fundamental limits on how much backend-specific information can be removed from classical outputs without degrading their usefulness. In addition, we observe that, for noisy quantum hardware, identifying fingerprints are inherently an intermediate-depth phenomenon, and establish a depth principle using Pauli-transfer-matrix tools. We complement the theory with experiments on Amazon Braket on AWS, using ion-trap and superconducting quantum processors. We observe 87-90% classification between superconducting backends and 96-100% classification across physical platforms, and find that identifiability can survive natural forms of post-processing. Overall, these results establish routing anonymity as a distinct security requirement for quantum cloud computing, and provide a framework for quantifying and controlling the utility-anonymity trade-off.
In shared quantum-cloud systems, variational programs encode costly classical optimization in rotation angles that can leak through adjacent hardware. We present the first physical recovery of a useful variational parameter region from a complete workload. The attack observes a disjoint idle two-qubit register during a 24-qubit quantum approximate optimization algorithm (QAOA) execution, without victim outputs or model queries. A fixed Marrakesh decoder distinguishes the regions under identical circuits, reaching 85/96 accuracy on held-out jobs and ROC AUC 0.997. Tests over separate time periods, matched controls, and a placement retest confirm the response at one of 17 layouts; conditional subsampling gives 0.756 expected accuracy from 2,000 observations. A public workload census finds source separation in 92 of 2,100 QAOA workloads, including five of 420 held-out workloads. For continuous secrets, a Jacobian factorization separates victim sensitivity, physical transfer, and scheduling, then converts channel gain into parameter and utility bounds. Under conditionally independent affine leakage, observable terms form a binary linear code. Its shortest element determines the first leakage order, while its dimension determines how many random bits perfect secrecy requires. Fresh masks defer leakage to the shortest cycle, while mask reuse creates a second-order temporal channel; the same code yields maximum safe slot concurrency.