問題文
Before deployment, a team runs sessions in which testers deliberately try to make an assistant produce harmful instructions. Which responsible AI principle does this activity serve most directly?
選択肢
- Privacy, because no personal data is used, and the sessions are run against a copy of the system.
- Explainability, because the sessions reveal how the model reaches its answers, and a system whose reasoning can be described is by definition safe to release.
- Safety, because the point is to find the conditions under which the system causes harm before real users reach them.
- Fairness, because testers come from different backgrounds, and a mixed group of testers is what makes the resulting judgment even-handed.