跳转至

大语言模型赋能的社交媒体机器人检测系统的攻击与防御

文章背景与核心概要

本文探讨了大语言模型（LLM）赋能的社交媒体机器人检测系统所面临的网络安全漏洞与防御机制。尽管大语言模型通过深度语义和上下文分析增强了威胁检测能力，但也带来了全新的攻击面。作者引入了两种新型对抗性攻击策略，利用这些语义弱点，将检测准确率降低了高达 48%。

为了缓解这一问题，研究团队提出了 LSABRE（LLM驱动的社交对抗性机器人识别集成系统），这是一个强大的多大语言模型框架。该框架能够保持检测的可靠性，在强大的对抗压力下仍维持 86% 的检测准确率。这项研究不仅局限于社交媒体机器人检测，其方法论和见解还可广泛应用于钓鱼检测、邮件分类和欺诈分析等多种LLM驱动的网络安全系统。

Breaking and Defending LLM-Powered Social Media Bot Detection Systems

Summary

The paper investigates the cybersecurity vulnerabilities and defense mechanisms of Large Language Model (LLM)-powered social media bot detection systems. While LLMs enhance threat detection through deep semantic and contextual analysis, they also expose new attack surfaces. The authors introduce two novel adversarial attack strategies that exploit these semantic weaknesses, degrading detection accuracy by up to 48%. To mitigate this, they propose LSABRE (LLM-powered Social Adversarial Bot Recognition Ensemble), a robust multi-LLM framework that preserves detection reliability, maintaining 86% detection accuracy under strong adversarial pressure.

元数据与出版详情

Metadata & Publication Details

  • arXiv ID: arXiv:2608.15893 [cs.AI]
  • 作者 (Authors): Nof Orenstein, Yoni Birman
  • 提交时间 (Submitted On): 2026年8月16日
  • 主学科 (Primary Subject): 人工智能 (cs.AI)
  • 录用/出版会议与期刊 (Accepted/Published Venues):
  • ACISP 2026 (澳大利亚信息安全与隐私会议)
  • IEEE 信息安全与隐私研讨会 (S&P) 2026 (海报展示)
  • Pragmatic Cybersecurity 2026, 1(2), 10 (DOI: 10.53941/pc.2026.100010)
  • arXiv ID: arXiv:2608.15893 [cs.AI]
  • Authors: Nof Orenstein, Yoni Birman
  • Submitted On: 16 August 2026
  • Primary Subject: Artificial Intelligence (cs.AI)
  • Accepted/Published Venues:
  • ACISP 2026 (Australasian Conference on Information Security and Privacy)
  • IEEE Symposium on Security and Privacy (S&P) 2026 (Poster)
  • Pragmatic Cybersecurity 2026, 1(2), 10 (DOI: 10.53941/pc.2026.100010)

摘要

Abstract

社交媒体机器人的崛起构成了持续的威胁，助长了虚假信息传播、舆论操纵以及对在线平台信任的侵蚀。为了打击这种现象，机器学习系统应运而生，用于检测和限制机器人活动，但攻击者通过对抗性学习和行为模仿等技术不断调整，加剧了机器人与检测工具之间持续的技术军备竞赛。

大语言模型（LLM）的最新进展通过对账户及其内容进行更深层次的语义和上下文分析，显著提升了机器人检测能力。然而，这种转变也引入了新的攻击面，使攻击者能够精心设计直接针对基于LLM的分类器的推理和生成机制的攻击手段。Anthropic 的 Claude Code Security 等行业工具同样利用 LLM 来做出安全关键决策，这进一步促使人们对其攻击面进行深入研究。

在这项工作中，我们研究了 LLM 赋能、特定威胁的网络安全应用中的进攻与防守两方面。虽然本研究聚焦于社交媒体机器人检测的挑战，但我们的方法论和见解可广泛推广至各类 LLM 赋能的网络安全系统，包括钓鱼检测、邮件分类和欺诈分析。

The rise of social media bots poses a persistent threat, enabling misinformation, opinion manipulation, and the erosion of trust in online platforms. To combat this, machine learning systems have been developed to detect and limit bot activity, but attackers continuously adapt through techniques such as adversarial learning and behavior imitation, fueling an ongoing arms race between bots and detection tools.

Recent advances in large language models (LLMs) have significantly improved bot detection by enabling deeper semantic and contextual analysis of accounts and their content. However, this shift also introduces new attack surfaces, allowing adversaries to craft exploits that directly target the reasoning and generation mechanisms of LLM-based classifiers. Industry tools such as Anthropic's Claude Code Security similarly leverage LLMs for security-critical decisions, further motivating a careful study of their attack surfaces.

In this work, we investigate both the offensive and defensive aspects of LLM-powered, threat-specific cybersecurity applications. While centered on the challenge of social media bot detection, our methodology and insights generalize to a broad class of LLM-powered cybersecurity systems, including phishing detection, email classification, and fraud analysis.

核心贡献：

Key Contributions:

  1. 进攻策略： 我们引入了两种新型对抗性攻击策略，系统性地利用了基于 LLM 的分类器的语义和上下文弱点，将其检测准确率降低了高达 48%。
  2. 防御架构： 为了应对这些威胁，我们提出了一种强大的多 LLM 防御架构，旨在自适应对抗条件下保持检测的可靠性。LSABRE（LLM-powered Social Adversarial Bot Recognition Ensemble，LLM驱动的社交对抗性机器人识别集成系统）是一个多 LLM 框架，它显着提升了一系列攻击下的鲁棒性，即使在强大且自适应的对抗压力下，也能保持 86% 的检测准确率。
  1. Offensive Strategies: We introduce two novel adversarial attack strategies that systematically exploit the semantic and contextual weaknesses of LLM-based classifiers, degrading their detection accuracy by up to 48%.
  2. Defensive Architecture: To counter these threats, we propose a robust multi-LLM defense architecture designed to preserve detection reliability under adaptive adversarial conditions. LSABRE (LLM-powered Social Adversarial Bot Recognition Ensemble) is a multi-LLM framework that substantially improves robustness across a range of attacks, maintaining 86% detection accuracy even under strong, adaptive adversarial pressure.

全文与资源

Full-Text & Resources

外部参考与工具

External References & Tools