Purpose
awesome-llm-security is a curated resource list for the field of large language model (LLM) security. It brings together tools, documents, and projects related to LLM security, helping researchers, developers, and security practitioners quickly understand and navigate the key materials in this field.
Main Capabilities
This repository organizes its content in list form, mainly divided into the following sections:
-
Papers: Contains research papers on LLM security, further categorized by topic, including:
- White-box attack
- Black-box attack
- Backdoor attack
- Fingerprinting
- Defense
- Platform Security
- Survey
- Benchmark
-
Tools: Lists multiple open-source tools for testing, evaluating, and enhancing LLM security, such as Plexiglass, PurpleLlama, Rebuff, Garak, LLMFuzzer, LLM Guard, Vigil, jailbreak-evaluation, Prompt Fuzzer, WhistleBlower, Open-Prompt-Injection, Agentic Radar, etc.
-
Articles: Provides reading materials on LLM security threats, attack surfaces, and defenses, such as Prompt Injection Cheat Sheet, OWASP Top 10 for LLM Applications, PoisonGPT, etc.
-
Other Awesome Projects: Includes LLM security startups, Adversarial Prompting, LLM security competitions, etc.
-
Other Useful Resources: Such as bug bounty programs (0din GenAI Bug Bounty), prompt injection games (Gandalf), CVE vulnerabilities, social media accounts, blogs, and newsletters.
How to Use
Users can directly visit the GitHub repository and browse categories of interest through the table of contents navigation. The repository provides links for each paper and tool (such as [paper], [repo], [site], etc.) for further reference. The repository also mentions sharing PDFs through Moonlight, providing paper abstracts and full texts.
Applicable Scenarios
- Academic research: Quickly find relevant papers and benchmarks in the field of LLM security.
- Security assessment: Obtain tools and frameworks for testing LLM security.
- Threat intelligence: Understand the latest attack methods, defense strategies, and security incidents.
- Education and training: Serve as a resource list for learning LLM security knowledge.
Limitations
- The repository itself is a resource list and does not provide direct security solutions or code implementations.
- The specific functions, effectiveness, pricing, etc. of the included papers and tools must be referenced through their respective links; this repository does not provide detailed information.
- Some resources may become outdated over time, so it is recommended to follow updates.
Contribution
The repository welcomes community contributions, and the Contribution Guidelines must be read before contributing.
