Question 400
A model endpoint outputs confidential training examples when prompted with generic cues. What observation would best confirm the intended security outcome?
Explanation
Correct Answer
A controlled safety test produces no prohibited real training records. This answer directly addresses the described constraint. Model behavior can expose sensitive memorized data. Successful validation requires: A controlled safety test produces no prohibited real training records.
Incorrect Answers
That approach concerns a different situation: a mobile app stores unencrypted session tokens in a world-readable local database. Mobile client storage is not inherently confidential on compromised devices. The required result is that a controlled safety test produces no prohibited real training records. The relevant evidence is: A controlled safety test produces no prohibited real training records.
It would address an app asks for contact-list access even though the business feature uses only a camera. The relevant reason in that case is: Unnecessary mobile permissions expand sensitive data exposure. Here the required success condition is a controlled safety test produces no prohibited real training records. The relevant evidence is: A controlled safety test produces no prohibited real training records.
This result applies to a different case: A security engineer needs to inspect BLE advertising data for an approved IoT sensor. Bluetooth advertisement fields can disclose device information passively. It does not establish the present requirement, where a controlled safety test produces no prohibited real training records. The relevant evidence is: A controlled safety test produces no prohibited real training records.