Quiz me — The LLM Threat Model

Q1.Match each attack surface to who typically controls the content that flows through it.

Q2.Select every statement that is true about delimiter-based mitigations (e.g. wrapping untrusted text in <<<DOC>>> tags).

Q3.Order these from most upstream (input) to most downstream (output/consumption) in a typical LLM request lifecycle.

Q4.Why don't classic mitigations like parameterized queries or input escaping solve prompt injection?

Q5.A risk's priority in a simple risk register is typically computed as likelihood times ______.

Q6.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?