CCAR-P: Claude Certified Architect — Professional

Question

During an architecture review, the security team identifies a risk that adversarial content inserted into retrieved documents could influence the model’s behavior.

Which mitigation most directly addresses this threat?

Explanation

Retrieved documents can carry indirect prompt-injection instructions. Treating all retrieved content as untrusted, applying classifiers or injection detection before it reaches the model, and validating outputs establishes controls at the relevant trust boundary.

Learn more

CCAR-P: Claude Certified Architect — Professional

AstroTutor

Notes

Readiness

QuestionQ83

Community Discussion