LLM Security
The field of security research and practices specifically for Large Language Models and generative AI.
LLM Security addresses security risks for language models: Prompt injection, jailbreaking, data leakage, model extraction. OWASP LLM Top 10 is the reference standard.
Explanation
LLM Security covers: Prompt injection, jailbreaking, data leakage, model extraction, training data poisoning. Differs from classic software security through natural language as attack surface.
Marketing Relevance
Every production AI needs a security concept: What data is in context? What actions can the AI perform? What happens if manipulated?
Example
OWASP LLM Top 10 documents: Prompt Injection (#1), Insecure Output Handling (#2), Training Data Poisoning (#3), Model Denial of Service (#4)...
Common Pitfalls
Classic security teams often don't understand LLM risks. New attack vectors emerge faster than defenses. Balance between security and usability.
Origin & History
LLM Security emerged as a field in 2022 with ChatGPT. Simon Willison coined "Prompt Injection". OWASP published LLM Top 10 (2023). Security conferences now have dedicated AI tracks.
Comparisons & Differences
LLM Security vs. AI Safety
AI Safety focuses on alignment and long-term risks; LLM Security on concrete attacks and defenses today.
LLM Security vs. Application Security
AppSec deals with code vulnerabilities; LLM Security deals with natural language as attack surface.