Security & AI Risk
OWASP publishes the first Top 10 for LLM Applications
The first community security taxonomy for LLM apps — prompt injection, sensitive-data leakage, insecure output handling — the baseline risk checklist for any clinical chatbot or copilot.