IEEE International Conference on Information Reuse and Integration (IRI)
Learning-Based Denial-of-Key-Generation (DoKG) Attack Against Quantum Key Distribution
Shradha Awasthi, Shafkat Islam, Md Tamjid Hossain
Abstract
Quantum Key Distribution (QKD) using the BB84 protocol enables secure communication based on quantum mechanics. While most security research addresses eavesdropping and key extraction attacks, denial-of-key generation (DoKG) attacks targeting the availability of key establishment remain under-explored. This paper presents a deep Q-network (DQN) based attacker that learns adaptive disruption strategies by exploiting classical-channel feedback in dynamic BB84 protocol. The attacker observes a six-dimensional state space derived from observable protocol parameters and selects from eleven discrete attack actions, including noisy gates, Pauli-X flips, attenuation, and strategic inactivity. The environment includes Markovian channel dynamics, adaptive photon intensity control by legitimate parties, and dynamically adjusted QBER thresholds. We evaluate the DQN approach against random and multi-armed bandit baselines across varying photon budgets and channel conditions. In the simulated BB84 environment, the DQN attacker achieved approximately 99% DoKG success while disrupting approximately 50.0-54.8% of photons, compared with 66.7-69.5% success and 58.1-59.1% photon disruption for the random baseline.