Why must you validate LLM output before using it in SQL, shell, or downstream APIs?

Insecure output handling

• Models can emit malicious SQL, shell commands, or JSON that bypasses your intent — classic OWASP LLM #2.