Question 1030
After implementing the design, the architecture team must validate: A developer proposes using AI model scanning to protect every runtime response, without request-path integration. What observed evidence best establishes the intended state?
Explanation
Correct Answer
Live prompts produce runtime enforcement logs; model scans remain associated with immutable artifact versions. It validates the recommended action: Design a separate inline AI runtime enforcement point for live requests while retaining predeploy scanning. The design principle is: Artifact scanning is not the same control location as runtime content evaluation.
Incorrect Answers
Synthetic malware and confidential data tests produce expected distinct prevention or DLP decisions. That observation supports a DIFFERENT situation: An AI workflow sends confidential documents to a third-party API without considering existing NGFW inspection. Here the selected design is: Design a separate inline AI runtime enforcement point for live requests while retaining predeploy scanning.
A lab agent's unauthorized purchase action is denied while approved read-only tool calls succeed. That observation supports a DIFFERENT situation: Autonomous procurement agents can call sensitive tools but their tool permissions were never bounded. Here the selected design is: Design a separate inline AI runtime enforcement point for live requests while retaining predeploy scanning.
The privacy review includes conversation IDs, log retention, user deletion processes and lawful processing purpose. That observation supports a DIFFERENT situation: An AI application claims its model is anonymous but retains conversation identifiers linked to employees. Here the selected design is: Design a separate inline AI runtime enforcement point for live requests while retaining predeploy scanning.