通过语义漂移检测，实时发现大模型中的隐蔽后门。
Detecting Sleeper Agents in Large Language Models via Semantic Drift Analysis
- 用句子嵌入分析语义偏差，结合测试题监控响应一致性。
- 在官方测试集上达92.5%准确率，零误报，<1秒/查询。
- 无需修改模型，适合部署阶段的安全审计与验证。
大型语言模型（LLMs）可能在特定部署条件下被植入恶意行为，训练时表现正常，这种现象称为“沉睡代理”。尽管近期研究显示此类后门能通过安全训练留存，但尚无实用检测方法。本文提出一种结合语义漂移分析与蜜罐基准对比的双方法检测系统，利用Sentence-BERT嵌入衡量与安全基线的语义偏离，并通过注入蜜罐问题监测响应一致性。在Cadenza-Labs提供的dolphin-llama3-8B沉睡代理模型上，系统实现92.5%准确率、100%精确率（零误报）和85%召回率。该方法可实时运行（单次查询<1秒），无需模型修改，是首个实用的LLM后门检测方案。本工作填补了AI部署中的关键安全空白，证明基于嵌入的检测可有效识别欺骗性行为而不影响部署效率。
原文摘要 · Abstract (English)
Large Language Models (LLMs) can be backdoored to exhibit malicious behavior under specific deployment conditions while appearing safe during training a phenomenon known as "sleeper agents." Recent work by Hubinger et al. demonstrated that these backdoors persist through safety training, yet no practical detection methods exist. We present a novel dual-method detection system combining semantic drift analysis with canary baseline comparison to identify backdoored LLMs in real-time. Our approach uses Sentence-BERT embeddings to measure semantic deviation from safe baselines, complemented by injected canary questions that monitor response consistency. Evaluated on the official Cadenza-Labs dolphin-llama3-8B sleeper agent model, our system achieves 92.5% accuracy with 100% precision (zero false positives) and 85% recall. The combined detection method operates in real-time (<1s per query), requires no model modification, and provides the first practical solution to LLM backdoor detection. Our work addresses a critical security gap in AI deployment and demonstrates that embedding-based detection can effectively identify deceptive model behavior without sacrificing deployment efficiency.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。