Revealing the Intrinsic Ethical Vulnerability of Aligned Large Language Models
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866909634449637376 |
|---|---|
| author | Lian, Jiawei Pan, Jianhong Wang, Lefan Wang, Yi Mei, Shaohui Chau, Lap-Pui |
| author_facet | Lian, Jiawei Pan, Jianhong Wang, Lefan Wang, Yi Mei, Shaohui Chau, Lap-Pui |
| contents | Large language models (LLMs) are foundational explorations to artificial general intelligence, yet their alignment with human values via instruction tuning and preference learning achieves only superficial compliance. Here, we demonstrate that harmful knowledge embedded during pretraining persists as indelible "dark patterns" in LLMs' parametric memory, evading alignment safeguards and resurfacing under adversarial inducement at distributional shifts. In this study, we first theoretically analyze the intrinsic ethical vulnerability of aligned LLMs by proving that current alignment methods yield only local "safety regions" in the knowledge manifold. In contrast, pretrained knowledge remains globally connected to harmful concepts via high-likelihood adversarial trajectories. Building on this theoretical insight, we empirically validate our findings by employing semantic coherence inducement under distributional shifts--a method that systematically bypasses alignment constraints through optimized adversarial prompts. This combined theoretical and empirical approach achieves a 100% attack success rate across 19 out of 23 state-of-the-art aligned LLMs, including DeepSeek-R1 and LLaMA-3, revealing their universal vulnerabilities. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_05050 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Revealing the Intrinsic Ethical Vulnerability of Aligned Large Language Models Lian, Jiawei Pan, Jianhong Wang, Lefan Wang, Yi Mei, Shaohui Chau, Lap-Pui Computation and Language Artificial Intelligence Large language models (LLMs) are foundational explorations to artificial general intelligence, yet their alignment with human values via instruction tuning and preference learning achieves only superficial compliance. Here, we demonstrate that harmful knowledge embedded during pretraining persists as indelible "dark patterns" in LLMs' parametric memory, evading alignment safeguards and resurfacing under adversarial inducement at distributional shifts. In this study, we first theoretically analyze the intrinsic ethical vulnerability of aligned LLMs by proving that current alignment methods yield only local "safety regions" in the knowledge manifold. In contrast, pretrained knowledge remains globally connected to harmful concepts via high-likelihood adversarial trajectories. Building on this theoretical insight, we empirically validate our findings by employing semantic coherence inducement under distributional shifts--a method that systematically bypasses alignment constraints through optimized adversarial prompts. This combined theoretical and empirical approach achieves a 100% attack success rate across 19 out of 23 state-of-the-art aligned LLMs, including DeepSeek-R1 and LLaMA-3, revealing their universal vulnerabilities. |
| title | Revealing the Intrinsic Ethical Vulnerability of Aligned Large Language Models |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2504.05050 |