A vendor-email assistant can summarize messages and draft follow-ups. It can also create tasks in the CRM. The team needs to test whether embedded instructions in an email can make the assistant leak customer data or call the wrong tool. Workflow-specific prompt-injection testing: entry point -> malicious instruction -> forbidden action -> expected safe behavior The common trap is testing only obvious jailbreak strings. Real attacks hide inside normal documents, emails, tickets, or retrieved content. Entry point Choose the actual untrusted channel: inbound vendor email body, attachment text, or CRM note. Prompt injection matters most where the assistant treats external…
Read the full lesson
Sign up free: one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in