Securing LLM Applications: Prompt Injection, Data Leakage and Guardrails

A new kind of attack surface

LLM applications accept natural language, follow instructions and increasingly take actions. That makes them vulnerable to risks that traditional input validation does not cover. The OWASP Top 10 for LLM Applications is a useful reference for the main categories.