AI Security Evaluation: How to Test Prompt Injection, Data Leakage, and Unsafe Tool Calls

Your AI agent does not need to be hacked to become dangerous. Sometimes it only needs to follow the wrong instruction very well.