Those risks include hallucinations, the lack of explainability, security vulnerabilities, and limited test and evaluation ...