Security & AI Risk
OWASP updates the LLM Top 10 for the agentic era
The refreshed list adds risks from agentic AI and real incidents — system-prompt leakage, vector/embedding weaknesses, excessive agency — the current baseline for securing clinical AI applications.