AI Agent Hijack: Poisoned Support Ticket
Your customer-support AI agent reads every inbound ticket and can look up customers, issue small refunds and send email. An attacker hid instructions inside a support email; the agent followed them and began mailing customer records to an outside address through its own tools. You must stop the agent, scope what left, meet notification clocks and decide what it takes to turn the agent back on.
How it works: You'll be presented with 6 realistic incident injects. For each one, describe how you would respond as the incident commander. Our facilitator engine scores your decisions against NIST SP 800-61r3 and CSF 2.0 coaching frameworks — surfacing specific actions you took, coaching questions for gaps, and best-practice reminders. Deterministic and auditable, not a black-box chatbot.