In short
There is far too much new malware to inspect by hand, so detection is automated with machine learning. Watching how a program behaves produces a huge number of features, which is hard for quantum models to take in.
Malware is growing faster than it can be checked
There were 450,000 to 560,000 new malware samples every day in 2025 (Afzalzada, 2026), and a significant portion (76%) of modern malware uses AI-driven polymorphism to evade modern signature-based detection techniques (House, 2026). This emphasises the need for faster, more scalable detection techniques that can handle the shifting landscape of malware.
Watching what a program does
Instead of looking for a known fingerprint, behavioural detection runs a program in a safe sandbox and records the API calls it makes: the requests it sends to the operating system, such as opening a file or connecting to the network. The pattern of calls says a lot about what the program is trying to do.
This research used the GBAC dataset (“Going Beyond API Calls in Dynamic Malware Analysis: A Novel Dataset”; Ilić et al., 2024) of 15,542 samples, for its nearly even split of benign and malware samples, 8,217 and 7,325 respectively. The dataset also has a high dimensionality, which directly instantiates the qubit encoding bottleneck: it becomes 262,144 features after hashing.
Why quantum, and what goes wrong
Quantum computing is an emerging technology that promises to be a prominent technology in computer security for identifying malware. Even with the development of more advanced quantum computing techniques such as Quantum Machine Learning (QML) and Variational Quantum Classifiers (VQC), a problem still presents itself: the number of qubits has a direct correlation to the size of the data that can be used for training. That is the qubit bottleneck.
Next: qubits and VQCs