Deep Reinforcement Learning (DRL) systems are increasingly used in safety-critical applications, yet their security remains severely underexplored. This work investigates backdoor attacks, which implant hidden triggers that cause malicious actions only when specific inputs appear in the observation space. Existing DRL backdoor research focuses solely on training-time attacks requiring full adversarial access to the training pipeline. In contrast, we reveal critical vulnerabilities across the DRL supply chain where backdoors can be embedded with significantly reduced adversarial privileges. We introduce two novel attacks: (1) TrojanentRL, which exploits component-level flaws to implant a persistent backdoor that survives full model retraining; and (2) InfrectroRL, a post-training backdoor attack which requires no access to training, validation, or test data. Empirical and analytical evaluations across six Atari environments show our attacks rival state-of-the-art training-time backdoor attacks while operating under much stricter adversarial constraints. We also demonstrate that InfrectroRL further evades two leading DRL backdoor defenses. These findings challenge the current research focus and highlight the urgent need for robust defenses.

With the rapid advancement of computing technologies, the frequency and complexity of cyber-attacks have escalated. Autonomous Cyber Defence (ACD) has emerged to combat these threats, aiming to train defensive agents that can autonomously respond to cyber incidents at machine speed and scale, similar to human defenders. One of the main challenges in ACD is enhancing the training efficiency of defensive agents in complex network environments, typically using Deep Reinforcement Learning (DRL). This work addresses this challenge by employing quantum-inspired methods. When coupled with Quantum-Inspired Experience Replay (QER) buffers and the Quantum Approximate Optimization Algorithm (QAOA), we demonstrate an improvement in training the defence agents against attacking agents in real-world scenarios. While QER and QAOA show great potential for enhancing agent performance, they introduce substantial computational demands and complexity, particularly during the training phase.