Detecting Sleeper Agents in Large Language Models via Semantic Drift Analysis
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866915627443159040 |
|---|---|
| author | Zanbaghi, Shahin Rostampour, Ryan Abid, Farhan Jarmakani, Salim Al |
| author_facet | Zanbaghi, Shahin Rostampour, Ryan Abid, Farhan Jarmakani, Salim Al |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2511_15992 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Detecting Sleeper Agents in Large Language Models via Semantic Drift Analysis Zanbaghi, Shahin Rostampour, Ryan Abid, Farhan Jarmakani, Salim Al Artificial Intelligence 68T50 I.2.6; I.2.7; K.6.5 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. |
| title | Detecting Sleeper Agents in Large Language Models via Semantic Drift Analysis |
| topic | Artificial Intelligence 68T50 I.2.6; I.2.7; K.6.5 |
| url | https://arxiv.org/abs/2511.15992 |