CY0-001 Practice Questions: AI-assisted security Domain

Correct answer (C): This is an indirect prompt injection attempt because untrusted email content is trying to control the AI copilot's behavior. The best mitigation is to enforce separation between instructions and untrusted content, so the copilot summarizes the message but does not follow embedded commands from the email body.

Why the other options are wrong:
- Option A: Confidence does not prove correctness and does not address malicious instructions embedded in input content.
- Option B: Blocking all copilot processing would reduce productivity more than necessary. The risk can be mitigated with instruction isolation and guardrails.
- Option D: Removing hidden text may help display issues, but relying on deletion alone is incomplete. The core control is preventing untrusted content from becoming executable instructions.

Correct answer (C): An AI agent connected to cloud control-plane APIs needs bounded permissions and governance controls. Scoped permissions, preapproved playbooks, human approval for high-impact actions, and complete logging preserve useful automation while reducing the risk of unauthorized, destructive, or poorly justified changes.

Why the other options are wrong:
- Option A: Explanations help reviewers, but they do not enforce authorization, prevent destructive actions, or create complete auditability.
- Option B: Read-only mode is safer but may unnecessarily eliminate approved low-risk automation. The question asks to preserve useful automation.
- Option D: AI confidence does not replace access control, approval gates, or rollback planning. High confidence can still accompany harmful remediation.

Correct answer (B): The proposed action is high impact because it could disrupt a critical payment processor and close alerts without full validation. AI-assisted SOAR should include human approval gates, scope validation, evidence review, escalation criteria, and rollback procedures before broad network blocks or alert closure actions execute.

Why the other options are wrong:
- Option A: A higher threshold may reduce some risky executions but still treats confidence as sufficient evidence. It does not address scope, approval, business impact, or rollback.
- Option C: Nonbusiness hours do not eliminate the risk of disrupting critical systems or closing alerts incorrectly. Timing is not a substitute for approval and validation.
- Option D: This increases automation risk and expands containment without malware confirmation. It may cause unnecessary disruption based on incomplete evidence.

Correct answer (B): The best next step is to validate the AI-generated CSPM finding against authoritative cloud configuration and data classification sources, then apply a controlled remediation that reduces public exposure. Deleting a production data container is a destructive action and should not be performed solely because an AI system recommended it.

Why the other options are wrong:
- Option A: A high AI confidence score can help prioritize investigation, but it is not proof that deletion is safe or required. This option risks operational disruption and data loss.
- Option C: Owner confirmation is useful, but closing the alert before validation could leave a public exposure unresolved.
- Option D: Sending potentially sensitive customer data to an external AI tool without approval creates a privacy and security risk.

Correct answer (B): The sandbox behavior conflicts with the benign AI verdict and includes suspicious actions such as encoded PowerShell, persistence, unknown domain communication, and credential access. The analyst should treat the file as suspicious, preserve evidence, and follow the required approval process for endpoint isolation or other high-impact containment.

Why the other options are wrong:
- Option A: This overtrusts the AI verdict. Confidence does not override contradictory sandbox indicators and unknown reputation.
- Option C: Running the file on another workstation is unsafe and unnecessary because the sandbox already showed suspicious behavior.
- Option D: Immediate deletion can destroy evidence and may not stop persistence or credential misuse. Containment should follow approved procedures.