Unique Security and Privacy Threats of Large Language Models: A Comprehensive Survey
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866912593317199872 |
|---|---|
| author | Wang, Shang Zhu, Tianqing Liu, Bo Ding, Ming Ye, Dayong Zhou, Wanlei Yu, Philip S. |
| author_facet | Wang, Shang Zhu, Tianqing Liu, Bo Ding, Ming Ye, Dayong Zhou, Wanlei Yu, Philip S. |
| contents | With the rapid development of artificial intelligence, large language models (LLMs) have made remarkable advancements in natural language processing. These models are trained on vast datasets to exhibit powerful language understanding and generation capabilities across various applications, including chatbots, and agents. However, LLMs have revealed a variety of privacy and security issues throughout their life cycle, drawing significant academic and industrial attention. Moreover, the risks faced by LLMs differ significantly from those encountered by traditional language models. Given that current surveys lack a clear taxonomy of unique threat models across diverse scenarios, we emphasize the unique privacy and security threats associated with four specific scenarios: pre-training, fine-tuning, deployment, and LLM-based agents. Addressing the characteristics of each risk, this survey outlines and analyzes potential countermeasures. Research on attack and defense situations can offer feasible research directions, enabling more areas to benefit from LLMs. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2406_07973 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Unique Security and Privacy Threats of Large Language Models: A Comprehensive Survey Wang, Shang Zhu, Tianqing Liu, Bo Ding, Ming Ye, Dayong Zhou, Wanlei Yu, Philip S. Cryptography and Security With the rapid development of artificial intelligence, large language models (LLMs) have made remarkable advancements in natural language processing. These models are trained on vast datasets to exhibit powerful language understanding and generation capabilities across various applications, including chatbots, and agents. However, LLMs have revealed a variety of privacy and security issues throughout their life cycle, drawing significant academic and industrial attention. Moreover, the risks faced by LLMs differ significantly from those encountered by traditional language models. Given that current surveys lack a clear taxonomy of unique threat models across diverse scenarios, we emphasize the unique privacy and security threats associated with four specific scenarios: pre-training, fine-tuning, deployment, and LLM-based agents. Addressing the characteristics of each risk, this survey outlines and analyzes potential countermeasures. Research on attack and defense situations can offer feasible research directions, enabling more areas to benefit from LLMs. |
| title | Unique Security and Privacy Threats of Large Language Models: A Comprehensive Survey |
| topic | Cryptography and Security |
| url | https://arxiv.org/abs/2406.07973 |