Policy Disruption in Reinforcement Learning:Adversarial Attack with Large Language Models and Critical State Identification

2025/07/24 by Junyong Jiang, Buwei Tian, Jiang, Junyong +7 · 1 citation
Computer Science · #Adversarial Robustness in Machine Learning #FOS: Computer and information sciences #Machine Learning (cs.LG)

paper · pdf · doi:10.48550/arxiv.2507.18113

openalex publication_date 2025/07/24 · openalex created_date 2025/10/16 · openalex updated_date 2026/07/28

Abstract

Reinforcement learning (RL) has achieved remarkable success in fields like robotics and autonomous driving, but adversarial attacks designed to mislead RL systems remain challenging. Existing approaches often rely on modifying the environment or policy, limiting their practicality. This paper proposes an adversarial attack method in which existing agents in the environment guide the target policy to output suboptimal actions without altering the environment. We propose a reward iteration optimization framework that leverages large language models (LLMs) to generate adversarial rewards explicitly tailored to the vulnerabilities of the target agent, thereby enhancing the effectiveness of inducing the target agent toward suboptimal decision-making. Additionally, a critical state identification algorithm is designed to pinpoint the target agent's most vulnerable states, where suboptimal behavior from the victim leads to significant degradation in overall performance. Experimental results in diverse environments demonstrate the superiority of our method over existing approaches.

Citations

• RAT: Adversarial Attacks on Deep Reinforcement Agents for Targeted Behaviors
• RLingua: Improving Reinforcement Learning Sample Efficiency in Robotic Manipulations With Large Language Models
• Efficient Adversarial Attacks on Online Multi-agent Reinforcement Learning
• Rethinking Adversarial Policies: A Generalized Attack Formulation and Provable Defense in RL
• Augmenting Reinforcement Learning with Transformer-based Scene Representation Learning for Decision-making of Autonomous Driving
• Understanding Adversarial Attacks on Observations in Deep Reinforcement Learning
• Real-time Adversarial Perturbations against Deep Reinforcement Learning Policies: Attacks and Defenses
• Robustifying Reinforcement Learning Agents via Action Space Adversarial Training
• Policy Teaching via Environment Poisoning: Training-time Adversarial Attacks against Reinforcement Learning
• Adaptive Reward-Poisoning Attacks against Reinforcement Learning
• Challenges and Countermeasures for Adversarial Attacks on Deep Reinforcement Learning
• A Review of Cooperative Multi-Agent Deep Reinforcement Learning
• Adversarial Policies: Attacking Deep Reinforcement Learning
• Emergent Complexity via Multi-Agent Competition
• Proximal Policy Optimization Algorithms
• Trust Region Policy Optimization

Cited by

• SIEVE: Selective Integrity Verification and Escalation for Defending LLM Agents against Indirect Prompt Injection

Related

• Optimal Attacks on Reinforcement Learning Policies
• Robust Deep Reinforcement Learning against Adversarial Behavior Manipulation
• Rethinking Adversarial Policies: A Generalized Attack Formulation and Provable Defense in RL
• Policy Teaching via Environment Poisoning: Training-time Adversarial Attacks against Reinforcement Learning
• Targeted Adversarial Attacks on Deep Reinforcement Learning Policies via Model Checking