Quantum machine learning assesses the resilience of post-quantum protocols

The Institute of Theoretical and Applied Informatics at the Polish Academy of Sciences is applying quantum machine learning in a new way: actively testing the defenses of existing protocols rather than creating new ciphers. This work details a successful demonstration of using quantum generative adversarial networks (QGANs) to load a probability distribution of hash-based digital signatures into the memory of a quantum computer. This represents a concrete step toward leveraging quantum computing to actively attack these signatures, even though there are limitations of short-term quantum-classical hybrid methods. The team sees this approach as the first step in a workflow for leveraging quantum computing to attack post-quantum cryptographic primitives, going beyond simply designing algorithms that are resistant to known quantum threats such as Scholl and Grover’s algorithms.

Post-quantum cryptography and quantum threat landscape

Quantum Generative Adversarial Networks (QGANs) are now being oriented toward actively testing post-quantum cryptographic defenses, rather than solely focusing on creating new cryptographic methods. This change means there is a growing recognition that today’s quantum computers can reveal vulnerabilities in algorithms designed to withstand future, more powerful machines. Researchers are no longer just interested in building unbreakable codes. They are actively investigating weaknesses in what is already proposed to be quantum-resistant. This achievement is currently not about deciphering signatures, but rather about establishing that capability and understanding how short-term quantum devices can be used in attacks. The study confirms that “short-term quantum-classical hybrid methods have the necessary capabilities for this purpose” and shows that even current techniques can be used to analyze and potentially compromise post-quantum schemes.

This is a move away from simply theorizing about the future threats posed by cryptography-related quantum computers (CRQC), which require large numbers of qubits, long coherence times, and high-fidelity quantum gates. Researchers demonstrate the application of QGAN to hash-based digital signatures and leverage QGAN to model the underlying probability distribution. This allows quantum computers to essentially “learn” the characteristics of signatures, potentially revealing patterns and weaknesses that traditional analysis misses. The team’s methodology builds on the fundamentals of classic generative adversarial networks (GANs), replacing components with quantum neural networks to exploit quantum phenomena such as superposition and entanglement. The ultimate goal is to ensure the long-term security of these protocols by identifying and addressing vulnerabilities before they can be exploited, and this work provides an important initial framework for doing so.

How the Shor & Grover algorithm affects public key systems

The field of cryptographic security is undergoing a fundamental shift, driven not by instant breaking of current encryption, but by the proactive assessment of vulnerabilities using new quantum tools. Although fully fault-tolerant quantum computers capable of cracking widely used public-key systems are still years away, researchers are now actively employing near-term quantum devices to probe the resiliency of post-quantum cryptography (PQC) protocols. Recent research has demonstrated surprising applications of quantum machine learning, specifically quantum generative adversarial networks (QGAN), not for creating new cryptography, but for rigorously testing existing PQC implementations. The researchers present an example application of QGAN to load probability distributions of hash-based digital signatures into the memory of a quantum computer, a concrete step toward exploiting quantum computers to attack these signatures. This is not about breaking signatures now, but establishing a workflow for future analysis.

The core of this approach lies in leveraging the unique capabilities of QGAN. A classic GAN includes two neural networks: a competitively trained generator and a discriminator. QGAN replaces these with quantum neural networks, offering the possibility of efficiently encoding complex probability distributions. This is very different from previous research. The researchers argue that this first step is critical to ensuring the long-term security of these protocols as quantum computing technology matures and the threat landscape evolves. The rapid adoption of the PQC standard, evidenced by its inclusion in the Java 26 and OpenSSH 10.0 suites released in March 2026, underscores the urgency of this research.

Jarosław A. Miszczak from the Institute of Theoretical and Applied Informatics of the Polish Academy of Sciences is researching new directions in quantum security research. Rather than focusing on building quantum computers powerful enough to break existing cryptography, his research investigates how even short-term machines can be used to test the resilience of proposed post-quantum cryptographic protocols. This represents a shift from a purely defensive strategy of designing algorithms to withstand future quantum attacks to a proactive, offensive approach focused on identifying vulnerabilities in current standards. This is not just a theoretical investigation. Researchers are actively working to model the behavior of cryptographic primitives on quantum hardware. This is especially important given the ongoing development and standardization of post-quantum cryptographic algorithms, such as the OpenSSH 10.0 suite and recent additions to Java, which are becoming increasingly popular in cloud environments and general-purpose programming languages. Java 26, released in March 2026, introduces support for post-quantum-ready JAR signing and hybrid public key cryptography to prepare your applications for the quantum era. The difference between the NISQ computer and its future counterpart, the Cryptography-Related Quantum Computer (CRQC), is important.

The field of cryptographic security is witnessing change. Rather than solely focusing on developing entirely new cryptographic systems, attention is now turning to actively testing the resilience of post-quantum protocols using the limited quantum computers available. The core of this strategy involves leveraging QGAN to model and analyze the probability distributions inherent in post-quantum cryptography primitives. QGAN translates this concept into the quantum realm, using parameterized quantum circuits to potentially enhance the generative power and efficiency of the process. Rather than immediately breaking post-quantum signatures, this research aims to establish the first steps in a workflow that will enable the use of quantum computing to attack post-quantum cryptographic primitives. By employing QGAN, the characteristics of signature schemes can be effectively “learned”, enabling targeted attacks and identifying potential vulnerabilities before they can be exploited by more powerful quantum computers. This example application using hash-based signatures suggests that QGAN is amenable to the analysis of other post-quantum cryptography schemes and could be a valuable tool for enhancing the security of future communication systems.

The assumption that quantum computing’s first impact on cryptography is about breaking codes is proving increasingly narrow. The focus has shifted to proactively evaluating the vulnerabilities of proposed post-quantum solutions, even given the limited quantum hardware currently available. The researchers aimed to investigate whether quantum machine learning could circumvent these protections. The presented approach can be used as the first step in a workflow that enables the use of quantum computing to attack post-quantum cryptographic primitives. This initial success paves the way for more advanced quantum-assisted attacks on post-quantum cryptography and will require continued evaluation and refinement of these emerging security standards.