Your platform team wants to catch prompt-injection, tool-abuse, and jailbreak risks in a multi-agent Foundry solution before production release. They already run unit and integration tests in CI. Which shift-left security practice best extends that pipeline for agentic threats?
Select an answer to reveal the explanation.
Short Explanation
Correct answer: D. Shift-left security for multi-agent systems means exercising adversarial AI risks—injection, goal hijacking, tool misuse—before release. Microsoft Foundry’s AI Red Teaming Agent is built for that style of automated probing and fits CI/CD gates alongside classic tests. Waiting for incidents (A) is reactive. Disabling tools (B) removes product value instead of securing it. SAST alone (C) does not model LLM behavioral failure modes. Wire red-team evaluations into promotion criteria with pass thresholds.
Full Explanation
Exam AI-500 expects candidates to implement shift-left security principles, including the AI Red Teaming Agent in Foundry. Traditional unit and integration tests validate deterministic code paths; they rarely explore prompt-injection chains, multi-agent collusion, or unsafe tool invocation sequences. Embedding automated red teaming in the evaluation stage of DTAP (or a canary pre-prod environment) surfaces those risks early. Fail the pipeline when severity exceeds policy. Option A postpones discovery to production. Option B is not a security strategy—it is feature removal. Option C addresses code vulnerabilities but not model or orchestration misuse. Combine red teaming with guardrail tests, least-privilege tool scopes, and content safety filters so findings drive concrete mitigations (prompt hardening, tool allowlists, output filters) rather than one-off manual reviews.