论文:叙事连贯的 LLM 蜜罐将高级攻击者会话时长从 40 提升到 165 条命令
Adaptive LLM-Driven Honeypots vs. LLM-Driven Attackers: Does Narrative Coherence Improve Deception?
做安全防御的工程师可以看看这篇。它用 MITRE Engage 场景库让蜜罐动态生成互相呼应的假文件和日志,把 LLM 攻击智能体的放弃率从 86% 压到 52%,推理成本还基本持平。如果你在搭钓鱼防御或评估自家蜜罐,这套按 ATT&CK 技术选场景的思路可以直接借鉴。
一篇 arXiv 论文提出"叙事欺骗"(narrative deception)防御策略:防御方 LLM 将攻击者每条命令映射到 MITRE ATT&CK 技术编号,再从基于 MITRE Engage 的场景库中选取兼容场景,在模拟 Ubuntu 主机上实时部署相互引用的文件、进程、服务和日志等假工件。
实验通过真实 SSH 会话让 LLM 防御方与自主攻击智能体对抗,共 1,150 个会话,对比 Static、Same-TTP、Multi-TTP、Chain 四档策略,覆盖 4 个开源攻击模型、4 个防御模型和 3 档攻击技能。在相同工件预算下,连贯叙事使高级攻击者的中位会话长度从 40 条命令增至 165 条,放弃率从 86% 降到 52%,效果随攻击技能提升而增强,且在全部 16 组模型配对中成立。
值得注意的是,几乎所有会话因攻击者"失去兴趣"而非识破欺骗而结束,作者据此建议蜜罐应持续制造有价值的发现,而不是只追求仿真保真度。
Adaptive LLM-Driven Honeypots vs. LLM-Driven Attackers: Does Narrative Coherence Improve Deception?
Large Language Models (LLMs) enable honeypots to generate responses and deceptive artefacts on demand, yet existing approaches evaluate generated content one reply at a time and leave open whether artefacts that jointly describe a coherent event retain an attacker better than unrelated ones. This question becomes pressing as autonomous LLM-driven offensive agents emerge as realistic adversaries. This paper proposes narrative deception, a reactive strategy in which the defender maps each observed attacker command to a MITRE ATT&CK technique, selects a compatible scenario from a curated catalogue based on MITRE Engage, and deploys mutually referential artefacts (files, processes, services, and logs) on an emulated Ubuntu host while the intrusion is under way. An LLM-powered defender and an autonomous LLM attacker interact through a real SSH session, and the attacker's termination policy distinguishes abandonment due to suspected deception from abandonment due to loss of interest. A campaign of 1,150 sessions compares four conditions of increasing sophistication (Static, Same-TTP, Multi-TTP, and Chain) across four open-weight attacker models, four defender models, and three attacker skill levels. Engagement increases significantly at every step. With the same artefact budget per reaction, coherent scenarios raise the median session length of advanced attackers from 40 to 165 commands and reduce their abandonment from 86% to 52%. The effect grows with attacker skill, holds across all sixteen model pairs, and comes at comparable inference cost. Nearly all sessions ended through loss of interest rather than detection, which suggests that deception should sustain meaningful findings, not only emulator fidelity.
本条情报归属 AITOP「论文」核心追踪栏目。在跨区域产业观察坐标中,该动态直接关联「AI 开发者工具链、Coding Agent 与工程化实践」研究主线,反映出当前技术演进与商业落地的最新风向。涉及核心实体与技术议题包括:LLM、蜜罐、MITRE ATT&CK、AI安全,读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。
信源核验说明:本文原始线索采编自「arXiv cs.AI」(@Pedro Beltrán-López, Raul Gonzalez Jimenez, Dario Maddaloni, Manuel Gil Pérez, Pantaleone Nespoli),经 AITOP 多维评分管线结构化提炼。