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

By Yifan Yao, Jinhao Duan, Kaidi Xu, Yuanfang Cai, Eric Sun, Yue Zhang. High Confid. Comput., 2023

1,265 citations recorded by the source corpus.

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.

Subfields: Artificial Intelligence, Sociology and Political Science

View Paper: https://www.semanticscholar.org/paper/383c598625110e0a4c60da4db10a838ef822fbcf

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Promises and challenges of generative artificial intelligence for human learning (2024, positive): “Although model alignment is often implemented to prevent GenAI from producing toxic content, recent evidence suggests that adversarial attacks using specific prompts can undermine these measures 85 .”
A Survey on the Memory Mechanism of Large Language Model-based Agents (2024, neutral): “Their memory can record cross-trial information through successful exemplars, which would be used as references on similar trials.”
Jailbreak and Guard Aligned Language Models With Only Few In-Context Demonstrations (2023, positive): “However, their widespread use has raised serious safety concerns [1], [2], particularly regarding their potential for generating harmful content (e.g., toxic, unethical, or illegal content).”
Security and Privacy Challenges of Large Language Models: A Survey (2024, neutral): “According to some recent studies [10] [194], the existing strategies can be used in the LLM context as well.”
Why Do Multi-Agent LLM Systems Fail? (2025, neutral): “These evaluations are crucial but primarily facilitate a top-down perspective, focusing on aggregate performance or high-level objectives like trustworthiness and security (Liu et al., 2023c; Yao et al., 2024b).”
Smaug: Fixing Failure Modes of Preference Optimisation with DPO-Positive (2024, methodology): “We evaluate the Llama-2 finetunes on MT-Bench [Zheng et al., 2023] This benchmark tests across multiple categories of LLM performance (for example: writing, roleplay, coding and math).”
Tool learning with large language models: a survey (2024, neutral): “Meanwhile, some studies highlight that LLMs can generate biased or harmful content and potentially leak sensitive information [193–195].”
Information Retrieval Using Large Language Models on a Restricted Access Environment (2026, unknown): “One concern is about confidential data [22], using external proprietary LLMs such as ChatGPT 2 , Grok 3 and Claude 4 requires sending the data to external servers, which may facilitate data leaks [23].”
Mobile Game Evaluation Method Based on Data Mining of Affective Time Series (2025, unknown): “Below, we describe the founding concepts of the method, which were jointly developed by the research team with the experts from the gaming company we worked with.”