RepreGuard: Detecting LLM-Generated Text by Revealing Hidden Representation Patterns
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866912541611917312 |
|---|---|
| author | Chen, Xin Wu, Junchao Yang, Shu Zhan, Runzhe Wu, Zeyu Luo, Ziyang Wang, Di Yang, Min Chao, Lidia S. Wong, Derek F. |
| author_facet | Chen, Xin Wu, Junchao Yang, Shu Zhan, Runzhe Wu, Zeyu Luo, Ziyang Wang, Di Yang, Min Chao, Lidia S. Wong, Derek F. |
| contents | Detecting content generated by large language models (LLMs) is crucial for preventing misuse and building trustworthy AI systems. Although existing detection methods perform well, their robustness in out-of-distribution (OOD) scenarios is still lacking. In this paper, we hypothesize that, compared to features used by existing detection methods, the internal representations of LLMs contain more comprehensive and raw features that can more effectively capture and distinguish the statistical pattern differences between LLM-generated texts (LGT) and human-written texts (HWT). We validated this hypothesis across different LLMs and observed significant differences in neural activation patterns when processing these two types of texts. Based on this, we propose RepreGuard, an efficient statistics-based detection method. Specifically, we first employ a surrogate model to collect representation of LGT and HWT, and extract the distinct activation feature that can better identify LGT. We can classify the text by calculating the projection score of the text representations along this feature direction and comparing with a precomputed threshold. Experimental results show that RepreGuard outperforms all baselines with average 94.92% AUROC on both in-distribution (ID) and OOD scenarios, while also demonstrating robust resilience to various text sizes and mainstream attacks. Data and code are publicly available at: https://github.com/NLP2CT/RepreGuard |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2508_13152 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | RepreGuard: Detecting LLM-Generated Text by Revealing Hidden Representation Patterns Chen, Xin Wu, Junchao Yang, Shu Zhan, Runzhe Wu, Zeyu Luo, Ziyang Wang, Di Yang, Min Chao, Lidia S. Wong, Derek F. Computation and Language Artificial Intelligence Detecting content generated by large language models (LLMs) is crucial for preventing misuse and building trustworthy AI systems. Although existing detection methods perform well, their robustness in out-of-distribution (OOD) scenarios is still lacking. In this paper, we hypothesize that, compared to features used by existing detection methods, the internal representations of LLMs contain more comprehensive and raw features that can more effectively capture and distinguish the statistical pattern differences between LLM-generated texts (LGT) and human-written texts (HWT). We validated this hypothesis across different LLMs and observed significant differences in neural activation patterns when processing these two types of texts. Based on this, we propose RepreGuard, an efficient statistics-based detection method. Specifically, we first employ a surrogate model to collect representation of LGT and HWT, and extract the distinct activation feature that can better identify LGT. We can classify the text by calculating the projection score of the text representations along this feature direction and comparing with a precomputed threshold. Experimental results show that RepreGuard outperforms all baselines with average 94.92% AUROC on both in-distribution (ID) and OOD scenarios, while also demonstrating robust resilience to various text sizes and mainstream attacks. Data and code are publicly available at: https://github.com/NLP2CT/RepreGuard |
| title | RepreGuard: Detecting LLM-Generated Text by Revealing Hidden Representation Patterns |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2508.13152 |