Technical practice
How would you validate and classify an AI runtime-defense alert before escalating it to L3?
What to coverPreparation suggestion: Explain how you would check the detection against prompt/response logs, decision traces, identity context and agent inventory, then assess severity and impact. Describe the case notes and reusable context you would pass to L3, including what remains uncertain.
How would you investigate a suspected prompt-injection or RAG-poisoning event?
What to coverPreparation suggestion: Distinguish an attacker-controlled instruction from ordinary source content, and trace how retrieved material influenced the model response or subsequent actions. Address evidence of sensitive-data leakage or guardrail failure without treating an alert alone as proof of compromise.
What would you check when an agent appears to use an unauthorized MCP tool?
What to coverPreparation suggestion: Compare the activity with the agent’s registered tools, owner, identity permissions and decision traces. Explain how you would distinguish an approved workflow from overprivileged or hijacked behavior and when you would escalate.
How can agent-inventory and ownership data improve runtime security decisions?
What to coverPreparation suggestion: Discuss how accurate ownership, MCP-registry entries and risk-scoring context help validate detections and support approval decisions. Consider what to do when an agent or tool has no clear owner or its inventory entry conflicts with observed activity.
How would you use KQL or SQL to turn AI-security telemetry into useful detection metrics?
What to coverPreparation suggestion: Describe the fields and correlations needed to link detections, agents, containment actions and case timestamps. Explain how you would define measures such as detection volume, containment rate and time to resolve, and guard against duplicates or missing data.
How would you report AI-security posture to operational leaders and auditors?
What to coverPreparation suggestion: Separate actionable operational trends from compliance evidence and risk summaries, stating the scope, time period and limitations of each measure. Discuss how documented cases and control evidence could support NIST AI RMF reporting, with EU AI Act–related evidence only where applicable.
How would you triage a cloud posture alert involving an exposed resource or excessive permissions?
What to coverPreparation suggestion: Explain how you would establish asset context, exposure, identity permissions and potential impact before routing, remediating or escalating under a playbook. Address how you would track remediation to closure and relate the finding to the stated CIS or NIST baselines without assuming every alert is exploitable.
How would you prioritize cloud workload vulnerabilities alongside Microsoft 365 or Google Workspace security signals?
What to coverPreparation suggestion: Compare workload exposure, vulnerability context and observed security signals rather than relying on severity labels alone. Describe the evidence needed to route each issue to the right team and record its disposition.
What would you automate with PowerShell or Python in an L2 investigation workflow?
What to coverPreparation suggestion: Propose a limited enrichment or reporting task using data the posting names, such as correlating alert identifiers with agent ownership or producing a metric. Explain validation, error handling and how an analyst would review results before using them in a case or report.
How would you evaluate a detection or policy update after an adversarial-test finding?
What to coverPreparation suggestion: Trace the stated loop from red-team finding to proposed rule or policy change and re-test validation. Discuss detection coverage, false positives and runbook changes, using OWASP LLM or agentic categories and MITRE ATLAS as investigation taxonomy where useful.
Behavioural practice
Tell me about a time you escalated an investigation that you could not resolve at your level.
What to coverPreparation suggestion: Use situation, action and result to show what you validated, why escalation was warranted and how you packaged evidence and open questions. Focus on making the handoff usable without claiming certainty you did not have.
Describe a time you handed over an active security case to another shift or team.
What to coverPreparation suggestion: Set out the situation, the notes and priorities you communicated, and the resulting continuity of work. Explain how you made outstanding actions, owners and timing clear across a distributed team.
Tell me about a time you turned security telemetry into a report for a nontechnical audience.
What to coverPreparation suggestion: Describe the situation, how you selected and checked the metrics, and what the audience could do with the result. Include how you conveyed uncertainty or data gaps rather than overstating the findings.
Describe a time you worked with another team to resolve a security finding.
What to coverPreparation suggestion: Give a situation, your actions with the relevant owners and the result, including how remediation was verified or tracked. This role works with the SOC, cloud teams, architects and incident responders, so make the coordination specific rather than describing teamwork in general.
Tell me about a time an investigation or test finding led you to improve a runbook or detection.
What to coverPreparation suggestion: Explain the original gap, your contribution to the change and how you checked its effect. Use situation, action and result, including any trade-off between catching more relevant activity and creating unnecessary alerts.