Parece simples até o entrevistador perguntar "why?". Abaixo, uma estrutura de resposta que dá conta do follow-up.
Controls around model inputs, outputs, and actions. Examples include data minimization, schema checks, content policy checks, authorization, tool allowlists, and resource budgets.
Treat injection detection as an imperfect signal. A prompt instruction or classifier is not an authorization boundary. Enforce permissions and allowed arguments in ordinary application code.
Define the user-visible fallback for each rejected or uncertain result, such as clarification, refusal, or human review. Measure false positives and false negatives rather than assuming a guardrail eliminates the risk.
Ler é o primeiro passo. No app, essa pergunta vira flashcard com repetição espaçada, e o módulo tem quiz e simulado para testar o que ficou.
Treinar essa pergunta no app