A company performs a final safety check before releasing a generative AI chatbot. What is the name of the evaluation method in which experts take an attacker's perspective and deliberately try to elicit harmful output or vulnerabilities to surface risks before deployment?
1 / 5
Select an answer