A is incorrect: PII detection is designed to identify and redact sensitive personal information, not to detect attempts at prompt leaking.

B is correct: Jailbreak detection, a component of Prompt Shields, is engineered to recognize and prevent efforts aimed at circumventing or exposing the AI system's underlying operational directives.

C is incorrect: Indirect attack detection focuses on identifying malicious instructions embedded in user-provided documents or external sources, rather than direct attempts to reveal the system prompt.

D is incorrect: Document translation is a functional capability for language conversion and has no role in detecting prompt leaking.