CCAR-P: Claude Certified Architect — Professional

Question

You are evaluating data-exfiltration risk in a Claude-based assistant that has tools for internal-document retrieval and outbound HTTP calls.

Which scenario most directly signals a data-exfiltration risk?

Explanation

Adversarial instructions embedded in retrieved content can redirect an assistant to invoke an outbound tool and transmit sensitive data to an attacker-controlled endpoint. This is an indirect prompt-injection path that enables unauthorized data exfiltration.

Learn more

Mitigate jailbreaks and prompt injections — Claude Platform Docs

CCAR-P: Claude Certified Architect — Professional

AstroTutor

Notes

Readiness

QuestionQ90

Community Discussion