LLM Security
This section covers the security of large language models themselves. It starts from prompt-engineering fundamentals and moves into the taxonomy, mechanics and bypass techniques of prompt injection, then walks through every risk in the OWASP LLM Top 10 together with its mitigations. Later articles cover the more engineering-facing topics: model supply chain, training-data poisoning and vector retrieval.
Everything follows the same structure: first why the risk exists, then the defensive actions that actually work — not just definitions.
This section currently holds the articles below, in suggested reading order.
-
Summary of Prompt Techniques
This section introduces basic prompt engineering techniques, primarily referencing Andrew Ng's prompt tutorial, as a precursor to prompt injection topics. This article do…
-
A Taxonomy of Prompt Injection
This post documents the various categories of Prompt Injection. Detailed techniques and tricks related to Prompt Injection will be covered in the next post, "A Compendium…
-
Prompt Injection: Cognitive Control
This is the first article in a series on Prompt Injection, with a total of five planned. The techniques recorded in this article number 68, primarily revolving around cog…
-
Prompt Injection via Boundary Manipulation
This is the second article in the Prompt Injection series, with a total of 5 planned. This article covers 5 techniques, focusing on boundary manipulation.
-
Prompt Injection: Instruction Restructuring
Instruction restructuring techniques are primarily designed to evade certain detection mechanisms. They encompass many methods, such as modifying how the attacker's instr…
-
Prompt Injection via Instruction Integration
This article focuses on instruction integration techniques in Prompt Injection. Instruction integration refers to scenarios where a malicious instruction sent by an attac…
-
Prompt Injection: Multimodal Attacks
Multimodal attacks essentially refer to delivering instructions in different modalities—such as audio, video, or images. These techniques can all be classified as multimo…
-
Prompt Injection: A Compilation of Cases
The related techniques about Prompt Injection have been basically summarized. If new techniques emerge in the future, they will be updated accordingly. This article will…
-
Concise Summary Table of Prompt Injection Techniques
Previously, I summarized various Prompt Injection techniques and comprehensive cases. However, due to the large number of techniques, it was difficult to master them all,…
-
Sensitive Information Leakage in LLM Security
Among the TOP 10 LLM security vulnerabilities, sensitive information leakage is included because AI systems are highly integrated with both internal and external data sou…
-
LLM Security: Insecure Plugins
One of the critical LLM vulnerability categories is Insecure Plugins. What exactly is an insecure plugin? LLMs have inherent limitations; many tasks require external tool…
-
LLM Security - Unrestricted Consumption
Previously, I published an article on DDoS attacks targeting LLMs, mainly focusing on network-layer DDoS aimed at disrupting services. This article focuses on unrestricte…
-
Data Poisoning in Large Model Security
Data poisoning refers to contaminating data so that it becomes malicious. For instance, during model training, training datasets scraped from the web may have been seeded…
-
LLM Security: Insecure Output
Among the vulnerabilities in large language models (LLMs), insecure output refers to the situation where the model produces hazardous content that can lead to undesirable…
-
Excessive Agency in LLM Security
When enterprises develop AI applications, they often build functionalities that are exposed to the large language model (LLM), such as operations on databases to retrieve…
-
LLM Security: DDOS Attacks
In the LLM OWASP Top 10, DDOS attacks were originally included but later removed during a revision. However, DDOS attacks are still frequently used. This article covers r…
-
LLM Security — Misinformation Vulnerability
In the context of LLM security, the misinformation issue was previously referred to as over-reliance — users depending on and trusting a model’s outputs, even when those…
-
LLM Supply Chain Vulnerabilities
A supply chain refers to the entire pipeline from raw material preparation, to production, to shipping, and finally to the recipient's usage. In the context of the intern…
-
Vector and Embedding Vulnerabilities in LLM Security
This issue specifically targets RAG (Retrieval-Augmented Generation) scenarios. RAG essentially equips a model with an external knowledge base. The typical pipeline invol…
-
Four Lines of Defense for LLM Security
We have previously systematically summarized the TOP10 vulnerabilities related to LLM security and conducted hands-on reproductions — you can refer to the earlier article…