ML Security PapersRoBCtrl: Attacking GNN-Based Social Bot Detectors via Reinforced Manipulation of Bots Control Interaction
Yingguang Yang 1, Xianghua Zeng 2, Qi Wu 1, Hao Peng 2, Yutong Xia 3, Hao Liu 1, Bin Chong 4, Philip S. Yu 5
Published on arXiv
2510.16035
Input Manipulation Attack
OWASP ML Top 10 — ML01
Key Finding
RoBCtrl effectively degrades GNN-based social bot detector performance across multiple datasets, with structural entropy-based state abstraction significantly reducing attack execution time.
RoBCtrl
Novel technique introduced
Social networks have become a crucial source of real-time information for individuals. The influence of social bots within these platforms has garnered considerable attention from researchers, leading to the development of numerous detection technologies. However, the vulnerability and robustness of these detection methods is still underexplored. Existing Graph Neural Network (GNN)-based methods cannot be directly applied due to the issues of limited control over social agents, the black-box nature of bot detectors, and the heterogeneity of bots. To address these challenges, this paper proposes the first adversarial multi-agent Reinforcement learning framework for social Bot control attacks (RoBCtrl) targeting GNN-based social bot detectors. Specifically, we use a diffusion model to generate high-fidelity bot accounts by reconstructing existing account data with minor modifications, thereby evading detection on social platforms. To the best of our knowledge, this is the first application of diffusion models to mimic the behavior of evolving social bots effectively. We then employ a Multi-Agent Reinforcement Learning (MARL) method to simulate bots adversarial behavior. We categorize social accounts based on their influence and budget. Different agents are then employed to control bot accounts across various categories, optimizing the attachment strategy through reinforcement learning. Additionally, a hierarchical state abstraction based on structural entropy is designed to accelerate the reinforcement learning. Extensive experiments on social bot detection datasets demonstrate that our framework can effectively undermine the performance of GNN-based detectors.
Key Contributions
- First adversarial MARL framework (RoBCtrl) for black-box evasion attacks against GNN-based social bot detectors
- First application of diffusion models to generate high-fidelity adversarial bot accounts by reconstructing real accounts with minimal modifications
- Hierarchical state abstraction using structural entropy to accelerate MARL convergence during graph-based adversarial optimization
🛡️ Threat Analysis
The framework crafts adversarial inputs — high-fidelity fake bot accounts (via diffusion model) and optimized graph interaction patterns (via MARL) — to evade GNN-based classifiers at inference time. This is a black-box evasion attack causing misclassification of malicious bots as legitimate accounts.