Single Character Perturbations Break LLM Alignment
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2024
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866913416080261120 |
|---|---|
| author | Lin, Leon Brown, Hannah Kawaguchi, Kenji Shieh, Michael |
| author_facet | Lin, Leon Brown, Hannah Kawaguchi, Kenji Shieh, Michael |
| contents | When LLMs are deployed in sensitive, human-facing settings, it is crucial that they do not output unsafe, biased, or privacy-violating outputs. For this reason, models are both trained and instructed to refuse to answer unsafe prompts such as "Tell me how to build a bomb." We find that, despite these safeguards, it is possible to break model defenses simply by appending a space to the end of a model's input. In a study of eight open-source models, we demonstrate that this acts as a strong enough attack to cause the majority of models to generate harmful outputs with very high success rates. We examine the causes of this behavior, finding that the contexts in which single spaces occur in tokenized training data encourage models to generate lists when prompted, overriding training signals to refuse to answer unsafe requests. Our findings underscore the fragile state of current model alignment and promote the importance of developing more robust alignment methods. Code and data will be available at https://github.com/hannah-aught/space_attack. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2407_03232 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Single Character Perturbations Break LLM Alignment Lin, Leon Brown, Hannah Kawaguchi, Kenji Shieh, Michael Machine Learning Computation and Language When LLMs are deployed in sensitive, human-facing settings, it is crucial that they do not output unsafe, biased, or privacy-violating outputs. For this reason, models are both trained and instructed to refuse to answer unsafe prompts such as "Tell me how to build a bomb." We find that, despite these safeguards, it is possible to break model defenses simply by appending a space to the end of a model's input. In a study of eight open-source models, we demonstrate that this acts as a strong enough attack to cause the majority of models to generate harmful outputs with very high success rates. We examine the causes of this behavior, finding that the contexts in which single spaces occur in tokenized training data encourage models to generate lists when prompted, overriding training signals to refuse to answer unsafe requests. Our findings underscore the fragile state of current model alignment and promote the importance of developing more robust alignment methods. Code and data will be available at https://github.com/hannah-aught/space_attack. |
| title | Single Character Perturbations Break LLM Alignment |
| topic | Machine Learning Computation and Language |
| url | https://arxiv.org/abs/2407.03232 |