Quantum Neural Networks (QNNs) represent a promising frontier in quantum machine learning, offering potential advantages in terms of expressivity and computational efficiency. However, their rapid development has outpaced critical attention to their security. In this work, we position QNN security as a foundational research challenge and present a comprehensive threat model spanning input encoding, circuit compilation, execution, and deployment. We introduce a taxonomy of emerging quantum-specific attack classes, including dataset poisoning, circuit tampering, side-channel leakage, multi-tenant interference, and federated learning threats. Our analysis highlights key gaps, including the lack of quantum-native robustness metrics, hardware variability, and cross-layer vulnerabilities. To address these issues, we outline strategic directions for developing secure-by-design QNN architectures, emphasizing formal verification, quantum-aware defenses, and adversarial benchmarking. This work lays the groundwork for a systematic approach to building trustworthy and resilient quantum learning systems.

Linear optical architectures have been extensively investigated for quantum computing and quantum machine learning applications. Recently, proposals for photonic quantum machine learning have combined linear optics with resource adaptivity, such as adaptive circuit reconfiguration, which promises to enhance expressivity and improve algorithm performances and scalability. Moreover, linear optical platforms preserve some subspaces due to the fixed number of particles during the computation, a property recently exploited to design a novel quantum convolutional neural networks. This last architecture has shown an advantage in terms of running time complexity and of the number of parameters needed with respect to other quantum neural network proposals. In this work, we design and experimentally implement the first photonic quantum convolutional neural network (PQCNN) architecture based on particle-number preserving circuits equipped with state injection, an approach recently proposed to increase the controllability of linear optical circuits. Subsequently, we experimentally validate the PQCNN for a binary image classification on a photonic platform utilizing a semiconductor quantum dot-based single-photon source and programmable integrated photonic interferometers comprising 8 and 12 modes. In order to investigate the scalability of the PQCNN design, we have performed numerical simulations on datasets of different sizes. We highlight the potential utility of a simple adaptive technique for a nonlinear Boson Sampling task, compatible with near-term quantum devices.

The Bell and Clauser-Horne-Shimony-Holt (CHSH) inequalities both test quantum mechanics against local hidden variable theories but differ fundamentally in experimental feasibility. The original Bell inequality requires perfect anti-correlation (identical outcomes when measurement settings align) and ideal detectors—assumptions unattainable in real-world setups due to noise and inefficiencies, limiting its practical utility. In contrast, the CHSH inequality tolerates imperfections, requiring only statistical correlation measurements across four setting combinations. This robustness makes it the preferred choice for experiments such as Mach-Zehnder interferometer-based tests, where quantum states (e.g., single photons) are manipulated to violate CHSH. While other Bell-type inequalities could theoretically be implemented in such setups, CHSH dominates due to its resilience to detector limitations and noise. These features make it pivotal for industry-scale Quantum Key Distribution (QKD) protocols. In QKD, entangled photon pairs are exchanged between Alice and Bob, who perform measurements in different bases, relying on perfectly randomly generated numbers. They calculate the CHSH value: S = \langle AB \rangle + \langle AB’ \rangle + \langle A’B \rangle – \langle A’B’ \rangle, where the angle brackets represent correlation functions between their measurement outcomes. If Alice and Bob observe S > 2, this confirms that they share genuine quantum entanglement; no eavesdropper can have complete information about their key, and the correlations cannot be explained by classical physics.

While the original Bell inequality’s mathematical rigour makes its practical adoption almost impossible, the CHSH approach is preferred not only for its experimental feasibility but also for its built-in self-testing capabilities, which are critical for real-world quantum security. This feature is essential for the mathematical formulation of QKD because it provides the theoretical foundation for device-independent CHSH security, the gold standard for quantum cryptography. Theoretically, it simultaneously verifies the quantum state, detects eavesdropping, enables scalable implementations, and maintains security even when the hardware cannot be trusted [1].

Despite the mathematical excellence of CHSH-based QKD, their engineering possibilities compromise their security foundations, creating space for the so-called Quantum Hacking. Several real-world quantum hacking techniques have been successfully demonstrated against commercial QKD systems as of 2025, utilising methods such as Phase Remapping Attack, Trojan Horse Attack, Time-Shift Attack, or Detector Blinding Attack. This work introduces a new method to exploit weak Random Number Generators (RNGs) in QKD systems using quantum RNG (qRNG) and machine learning. By monitoring entropy and external factors (e.g., temperature), we demonstrate how compromised RNGs enable side-channel attacks. Our framework integrates:
– Multi-modal RNG Analysis: A 12-qubit qGAN quantifies distribution similarity via relative entropy (KL divergence <0.1) and discriminator loss (3.7–17), validated on Rigetti Aspen-M-3 (80 qubits) and IonQ Aria-1 (25 qubits) hardware.
– Bias Detection: Markov chain-enhanced logistic regression identifies hardware-induced biases (59% ‘1’ frequency in compromised RNGs vs. 54% in certified devices).
– Real-Time Monitoring: A neural network classifier (58.7% accuracy) discriminates quantum-generated randomness from classical simulations using 100-bit entropy profiles.

Experimental results show CHSH scores correlate with RNG quality (Rigetti: 0.8036, IonQ: 0.8362), while gate fidelity (IonQ: 99.4% vs. Rigetti: 93.6%) impacts certifiable randomness. Combining device-independent CHSH validation with machine learning, this framework detects attacks like phase remapping and detector blinding through entropy deviations.

By bridging theoretical security with engineering realities, our methodology addresses critical QKD implementation gaps. For instance, temperature-induced RNG biases—tracked via real-time entropy monitoring—enable side-channel exploitation unless mitigated by adaptive protocols. This approach enhances NIST’s « randomness extractor » pipeline, ensuring compliance with post-quantum standards under realistic noise conditions [2].

References:

[1] Zapatero, V., van Leent, T., Arnon-Friedman, R. et al. Advances in device-independent quantum key distribution. npj Quantum Inf 9, 10 (2023).

https://www.nature.com/articles/s41534-023-00684-x

[2] Kołcz, H., Pandey, T., Shah, Y. et al. Verification of qRNG with qGAN and Classification Models. QUACC+ CTP PAS, Warsaw, Poland (2025).

Poster: https://github.com/hubertkolcz/NoiseVsRandomness

BOA: https://quacc2025.cft.edu.pl/QUACC+2025_BOA.pdf

The transition of quantum computing into practical applications depends on successfully combining quantum algorithms with existing classical software systems. Resources for creating hybrid applications are currently fragmented which demands teamwork efforts across various disciplines. Moreover there currently are no clear predefined methods, standards or solutions for these complex operations. We introduce hereby EniQmA which stands as a quantum software engineering framework for unifying this development process as well as deployment and operation of hybrid quantum-classical applications. EniQmA integrates agile workflows with domain-specific toolchains along with quantum DevOps practices to enable connecting theoretical quantum algorithms with scalable industrial applications. The framework delivers structured process models together with quality standards and cohesive dvelopment tools to produce reliable and sustainable quantum software. Through its low-code/no-code capabilities EniQmA makes quantum application development accessible to domain experts who have minimal quantum expertise or enable specialists with no programming experience. Our framework demonstrates its effectiveness in three real-world industrial scenarios which show how it helps move research into practical deployment. EniQmA creates an essential initial step for standardizing quantum software engineering which speeds up the journey to practical quantum advantage in industrial applications.

We propose a new scheme for near-term photonic quantum devices that allows to increase the expressive power of the quantum models beyond what linear optics can do. This scheme relies upon state injection, a measurement-based technique that can produce states that are more controllable, and solve learning tasks that are believed to be intractable classically. We explain how circuits made of linear optical architectures separated by state injections are well-suited for experimental implementation. In addition, we give theoretical results regarding the evolution of the purity of the resulting states, and we discuss how it impacts the distinguishability of the circuit outputs. Finally, we study a computational subroutine of learning algorithms named probability estimation, and we show that the state injection scheme we propose may offer a potential quantum advantage in a regime that can be more easily achieved than state-of-the-art adaptive techniques. Our analysis offers new possibilities for near-term advantage that rely on overcoming fewer experimental difficulties.