Beyond Training-time Poisoning: Component-level and Post-training Backdoors in Deep Reinforcement Learning
This work addresses security vulnerabilities in DRL systems for safety-critical applications, revealing critical supply chain flaws that are more realistic than prior attacks.
The paper tackles the problem of backdoor attacks in deep reinforcement learning by introducing two novel attacks, TrojanentRL and InfrectroRL, which operate with reduced adversarial privileges compared to existing training-time methods, achieving performance rivaling state-of-the-art attacks across six Atari environments.
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 unrealistic 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, nor 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.