Quiz me — The LLM Threat Model

Q1.Order these from most upstream (input) to most downstream (output/consumption) in a typical LLM request lifecycle.

Q2.A risk's priority in a simple risk register is typically computed as likelihood times ______.

Q3.Why don't classic mitigations like parameterized queries or input escaping solve prompt injection?

Q4.Match each attack surface to who typically controls the content that flows through it.

Q5.In the OWASP LLM Top 10, which category covers a model being tricked into invoking a tool or taking an action beyond its intended scope?

Q6.Select every statement that is true about delimiter-based mitigations (e.g. wrapping untrusted text in <<<DOC>>> tags).