[Paper Review] A Survey on Large Language Model (LLM) Security and Privacy: The Good, the Bad, and the Ugly

Yifan Yao, Jinhao Duan+4 more|arXiv (Cornell University)|Dec 4, 2023

Privacy-Preserving Technologies in DataComputer Science313 references44 citations

TL;DR

A comprehensive literature review categorizing LLM security and privacy work into three parts—positive security benefits, potential attack pathways, and inherent vulnerabilities with defenses.

ABSTRACT

Large Language Models (LLMs), such as ChatGPT and Bard, have revolutionized natural language understanding and generation. They possess deep language comprehension, human-like text generation capabilities, contextual awareness, and robust problem-solving skills, making them invaluable in various domains (e.g., search engines, customer support, translation). In the meantime, LLMs have also gained traction in the security community, revealing security vulnerabilities and showcasing their potential in security-related tasks. This paper explores the intersection of LLMs with security and privacy. Specifically, we investigate how LLMs positively impact security and privacy, potential risks and threats associated with their use, and inherent vulnerabilities within LLMs. Through a comprehensive literature review, the paper categorizes the papers into "The Good" (beneficial LLM applications), "The Bad" (offensive applications), and "The Ugly" (vulnerabilities of LLMs and their defenses). We have some interesting findings. For example, LLMs have proven to enhance code security (code vulnerability detection) and data privacy (data confidentiality protection), outperforming traditional methods. However, they can also be harnessed for various attacks (particularly user-level attacks) due to their human-like reasoning abilities. We have identified areas that require further research efforts. For example, Research on model and parameter extraction attacks is limited and often theoretical, hindered by LLM parameter scale and confidentiality. Safe instruction tuning, a recent development, requires more exploration. We hope that our work can shed light on the LLMs' potential to both bolster and jeopardize cybersecurity.

Motivation & Objective

• Assess how LLMs positively impact security and privacy across domains.
• Identify risks and threats arising from LLM usage in cybersecurity.
• Examine intrinsic vulnerabilities and defenses of LLMs across training and inference stages.

Proposed method

• Conduct a comprehensive literature review of 281 papers on LLM security and privacy.
• Categorize papers into The Good (security benefits), The Bad (offensive uses), and The Ugly (vulnerabilities and defenses).
• Synthesize findings to compare LLM-based methods with state-of-the-art approaches across domains.

Experimental results

Research questions

• RQ1 RQ1: How do LLMs positively contribute to security and privacy across diverse domains, and what advantages do they offer to the security community?
• RQ2 RQ2: What potential risks and threats emerge from the utilization of LLMs within cybersecurity?
• RQ3 RQ3: What vulnerabilities and weaknesses within LLMs, and how to defend against those threats?

Key findings

• LLMs predominantly enhance security, especially in code security and data security/privacy, often outperforming traditional methods.
• User-level attacks are the most prevalent due to LLMs’ human-like reasoning abilities, posing security and privacy risks.
• Vulnerabilities in LLMs fall into AI model inherent and non-AI inherent categories, with defenses spanning architecture, training, and inference phases.
• Safe instruction tuning and model extraction research are still limited and largely theoretical given the scale and confidentiality of LLMs.
• The study is the first to comprehensively cover good, bad, and ugly aspects of LLM security and privacy in a unified survey.